Cambridge Commentary on EU General-Purpose AI Law

Chapter VII
AI Office
Commentary by Gregor Gindlin & Matt Hanrahan (joint first authors)

AI Act provision

Article 64

  1. The Commission shall develop Union expertise and capabilities in the field of AI through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .
  2. Member States shall facilitate the tasks entrusted to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , as reflected in this Regulation.

Recitals

Recital 148

This Regulation should establish a governance framework that both allows to coordinate and support the application of this Regulation at national level, as well as build capabilities at Union level and integrate stakeholders in the field of AI. The effective implementation and enforcement of this Regulation require a governance framework that allows to coordinate and build up central expertise at Union level. The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. was established by Commission Decision (45) and has as its mission to develop Union expertise and capabilities in the field of AI and to contribute to the implementation of Union law on AI. Member States should facilitate the tasks of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. with a view to support the development of Union expertise and capabilities at Union level and to strengthen the functioning of the digital single market. Furthermore, a Board composed of representatives of the Member States, a scientific panel to integrate the scientific community and an advisory forum to contribute stakeholder input to the implementation of this Regulation, at Union and national level, should be established. The development of Union expertise and capabilities should also include making use of existing resources and expertise, in particular through synergies with structures built up in the context of the Union level enforcement of other law and synergies with related initiatives at Union level, such as the EuroHPC Joint Undertaking and the AI testing and experimentation facilities under the Digital Europe Programme.

(45) Commission Decision of 24.1.2024 establishing the European Artificial Intelligence Office C(2024) 390.

Recital 162

To make best use of the centralised Union expertise and synergies at Union level, the powers of supervision and enforcement of the obligations on providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. should be a competence of the Commission. The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. should be able to carry out all necessary actions to monitor the effective implementation of this Regulation as regards general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. . It should be able to investigate possible infringements of the rules on providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. both on its own initiative, following the results of its monitoring activities, or upon request from market surveillance authorities Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. in line with the conditions set out in this Regulation. To support effective monitoring of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , it should provide for the possibility that downstream providers Article 3(68) AI Act: ‘downstream provider’ means a provider of an AI system, including a general-purpose AI system, which integrates an AI model, regardless of whether the AI model is provided by themselves and vertically integrated or provided by another entity based on contractual relations. lodge complaints about possible infringements of the rules on providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and systems.

Recital 164

The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. should be able to take the necessary actions to monitor the effective implementation of and compliance with the obligations for providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. laid down in this Regulation. The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. should be able to investigate possible infringements in accordance with the powers provided for in this Regulation, including by requesting documentation and information, by conducting evaluations, as well as by requesting measures from providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. . When conducting evaluations, in order to make use of independent expertise, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. should be able to involve independent experts to carry out the evaluations on its behalf. Compliance with the obligations should be enforceable, inter alia, through requests to take appropriate measures, including risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. mitigation measures in the case of identified systemic risks Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. as well as restricting the making available on the market Article 3(10) AI Act: ‘making available on the market’ means the supply of an AI system or a general-purpose AI model for distribution or use on the Union market in the course of a commercial activity, whether in return for payment or free of charge. , withdrawing or recalling the model. As a safeguard, where needed beyond the procedural rights provided for in this Regulation, providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. should have the procedural rights provided for in Article 18 of Regulation (EU) 2019/1020, which should apply mutatis mutandis, without prejudice to more specific procedural rights provided for by this Regulation.

Select bibliography

  • Herbers B and Rappenglück D, ‘Die Durchsetzung der KI-Verordnung auf EU-Ebene: Das EU AI Office’ (2024) Recht Digital 432.
  • Hilgendorf E and Härtlein J, ‘Art. 64 Büro für Künstliche Intelligenz’ in Eric Hilgendorf and Johannes Härtlein (eds.), KI-VO: Verordnung über künstliche Intelligenz (Nomos 2025).
  • Novelli C, Hacker P, Morley J, Trondal J and Floridi L, ‘A Robust Governance for the AI Act: AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , AI Board, Scientific Panel, and National Authorities’ (2025) 16 European Journal of Risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. Regulation 566.
  • Paschke A and Rachut S, ‘Art. 64 Büro für Künstliche Intelligenz’ in Jens Schefzig and Robert Kilian (eds), Beck’scher Online-Kommentar KI-Recht (4th edn, C H Beck 2025).
  • Roth-Isigkeit D, ‘Art. 64 Büro für Künstliche Intelligenz’ in Mario Martini and Christiane Wendehorst (eds), KI-VO: Verordnung über Künstliche Intelligenz: Kommentar (2nd edn, C H Beck 2026).
  • Schmidl M and Rohner A, ‘Article 64 AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ in Ceyhun Necati Pehlivan, Nikolaus Forgó and Peggy Valcke (eds), The EU Artificial Intelligence (AI) Act: A Commentary (Wolters Kluwer 2024).
  • Wende S, ‘Art. 64 Büro für Künstliche Intelligenz’ in David Bomhard, Fritz-Ulli Pieper and Susanne Wende (eds), Kommentar KI-VO: Verordnung über Künstliche Intelligenz (Fachmedien Recht und Wirtschaft 2025).
  • Wendt J and Wendt D, ‘§ 13 Aufsichtssystem’ in Janine Wendt and Dominik Wendt (eds), Das neue Recht der Künstlichen Intelligenz (Nomos 2025).

Commentary

1. General remarks

1.1. Introduction

1Article 64 AI Act1 is entitled ‘ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ and concerns the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s mission and its relationship with the Commission and Member States.2 The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. – short for European Artificial Intelligence Office3 – is part of the Commission’s administrative structure as Directorate A of the Directorate-General for Communication Networks, Content and Technology (“DG CNECT”)4 and plays a key role in the governance of AI at the EU level.5 In particular, it carries out many relevant tasks concerning the governance of general-purpose AI (“GPAI”) models,6 supporting the Commission’s monitoring, supervision and enforcement of corresponding GPAI model obligations.7

2The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. was established under Article 1(1) of the Commission Decision of January 2024 establishing the European Artificial Intelligence Office (C/2024/1459) (“Establishment Decision”)8 – in anticipation of the then forthcoming AI Act.9 By establishing the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. before the adoption of the AI Act, the Commission sought to ‘allow for the preparation of the implementation of the [AI Act] to start as soon as possible’.10 As evidenced by the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s nature as a Commission directorate,11 the European Parliament’s proposal to establish the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as an independent body of the Union with legal personality12 did not survive the trilogue negotiations.13 Interestingly, the Commission’s original draft AI Act dated April 2021 did not provide for an AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. at all.14

3Provisions on the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s establishment, nature and functions are distributed across both the Establishment Decision and the AI Act.15 In particular, Article 1 of the Establishment Decision concerns the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s establishment, whereas Articles 2 and 3 concern its mission and tasks. In the AI Act, Articles 3(47), 64 and 88(1) are of particular relevance to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. and its role.16 Article 3(47) defines the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as ‘the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. and general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. , and AI governance, provided for in [the Establishment Decision]’. This article also provides that the AI Act’s references to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall be construed as references to the Commission.17

4Article 64, located in the AI Act’s chapter on ‘Governance’,18 complements this definition by outlining an aspect of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s mission in its first paragraph19 and requiring Member States to support the work of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. in its second paragraph.20 Article 88(1), which opens the AI Act’s section on ‘Supervision, investigation, enforcement and monitoring in respect of providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. 21, further supplements these provisions by stating that the Commission ‘shall entrust the implementation’ of its supervisory and enforcement tasks to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .22 Numerous further provisions of the AI Act refer to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. in various contexts.23

5The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is part of the AI Act’s larger governance framework and needs to be distinguished from the other Union-level governance entities referred to in Section 1. of Chapter VII of the AI Act,24 namely the European Artificial Intelligence Board (“Board”),25 the advisory forum,26 and the scientific panel of independent experts (“scientific panel”).27 The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. plays a central role in this governance framework,28 exercising coordination functions and providing administrative support.29 Among other tasks, the Board, the advisory forum and the scientific panel support the Commission and the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. variously in the exercise of their respective tasks under the AI Act.30

1.2. Structure & overview

6This chapter examines the role of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. under the AI Act. Section 2. begins with a textual analysis of Article 64, before Section 3. addresses the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s institutional status. This latter section analyses the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role as a Commission directorate31 and its lack of legal personality,32 while also discussing whether it has a special institutional status33 and the relevance of its denomination as an ‘office’ with regard to judicial review under the Treaties.34 The analysis further establishes that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is not a European office under Regulation (EU, Euratom) 2024/2509 (“Financial Regulation”).35 Section 4. then considers the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s remit, discussing both its role in the AI Act’s implementation36 and its further tasks beyond the AI Act’s scope.37 Section 5. briefly describes the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s functioning, before the chapter concludes in Section 6. with an evaluation of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s functioning foreseen by the AI Act.

7As may have become apparent already from the foregoing, Article 64 is closely linked to various other provisions within the AI Act, particularly Article 3(47), which defines the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , and Article 88,38 which addresses the supervision and enforcement of obligations on GPAI model providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. , and cannot be interpreted in isolation. Where relevant, the following analysis therefore provides cross-references to other chapters of this Commentary examining those other provisions and the interpretive issues they raise.

2. Article 64

8Article 64 is the shortest article in Section 1. entitled ‘Governance at Union level’ of Chapter VII of the AI Act. Its first paragraph concerns the Commission’s development of Union AI expertise and capabilities,39 while its second paragraph concerns the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s relationship with the Member States.40 Important aspects concerning the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , including its establishment, legal nature and remit are found in other provisions of the AI Act and the Establishment Decision.41

2.1. Article 64(1): Union expertise and capabilities

9The first paragraph of Article 64 declares that the Commission ‘shall develop Union expertise and capabilities in the field of AI through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ (emphasis added).42 Both the AI Act and the Establishment Decision regard Union expertise and capabilities as a necessary basis for the effective implementation and enforcement of the AI Act.43 The Commission’s responsibility under Article 64(1) to build Union AI expertise and capabilities via the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. aligns with the EU’s broader Digital Europe Programme, an EU-run funding programme with an overall budget exceeding €8 billion that focuses on ‘support[ing] and accelerat[ing] the digital transformation of the European economy, industry and society’.44 One of the five specific objectives of that programme centres on artificial intelligence and includes ‘build[ing] up and strengthen[ing] core AI capacities and knowledge in the Union’.45 In particular, the Digital Europe Programme also contributes to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s funding.46

10It is possible to conceive of different ways through which the Commission can develop AI expertise47 and capabilities48, including – as Recital 148 notes – by ‘making use of existing resources and expertise’.49 This encompasses utilising ‘synergies with structures built up in the context of the Union level enforcement of other law and synergies with related initiatives at Union level, such as the EuroHPC Joint Undertaking and the AI testing and experimentation facilities under the Digital Europe Programme’.50 It is noteworthy, however, that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s own remit indisputably extends beyond expertise development and capability building,51 encompassing direct participation in implementation, supervision, monitoring, investigation and enforcement responsibilities concerning GPAI models.52

2.2. Article 64(2): Relationship with Member States

11Article 64(2) provides that Member States ‘shall facilitate the tasks entrusted to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’.53 This duty is reminiscent of the Member States’ duty of sincere cooperation under the Treaties,54 and resembles the third subparagraph of Article 4(3) of the Treaty on European Union55 (“TEU”): ‘The Member States shall facilitate the achievement of the Union’s tasks and refrain from any measure which could jeopardise the attainment of the Union’s objectives.’56

12Article 64(2)’s duty to facilitate likely extends to all tasks of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. under the AI Act. Given the unclear reference point of the addition ‘as reflected in this Regulation’ in Article 64(2), one might read this as limiting the provision’s scope to instances where the AI Act already expressly requires Member States to support the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. in any form.57 However, this reading would render Article 64(2) largely redundant – in tension with the principle that where a provision of EU law is open to several interpretations, preference must be given to the interpretation which ensures that it is not rendered redundant and, thus, retains its effectiveness.58 The phrase ‘as reflected in this Regulation’ should therefore be understood as referring to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s tasks generally pursuant to the AI Act, while clarifying that Article 64(2) does not extend to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s tasks outside the AI Act.59 The duty of sincere cooperation under Article 4(3) TEU would, in any event, apply to such extraneous tasks.

13Article 64(2) is not a hollow provision. Indeed, failure of a Member State to comply with its duty of cooperation under Article 64(2) may result in the Commission initiating infringement proceedings pursuant to Article 258 TFEU.60

14Notably, the duty of cooperation is not unilateral in favour of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. . The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s duty to cooperate with authorities and bodies of the Member States in the performance of its tasks under Article 2(3) of the Establishment Decision constitutes a counterpart to the Member States’ duty under Article 64(2).61

3. Institutional status

3.1. Commission directorate

15The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. forms part of the Commission as Directorate A of the Directorate-General for Communication Networks, Content and Technology (“DG CNECT”).62 Directorates-General, including their directorates63, assist the Commission in preparing and performing its tasks and achieving its priorities.64 The nature of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as a Commission directorate is reflected in the definition of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. found in Article 3(47),65 according to which (i) the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is a ‘function’ of the Commission and (ii) references in the AI Act to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. must be construed as references to the Commission. The Commission Guidelines set out that the Commission ‘act[s] through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ when supervising and enforcing the obligations for GPAI model providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. under the AI Act.66

16Corresponding to its status as a Commission directorate, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is subject to DG CNECT’s annual management plan67 and operates ‘in accordance with Commission internal processes’.68 The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is also subject to the Commission’s supervision,69 with the Commission being legally responsible for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s actions.70 Some authors have concluded, particularly based on the language of Article 3(47), that entities subject to the AI Act do not have a subjective right to have their affairs handled by the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. specifically.71 For officials working in the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , human resources are drawn from within DG CNECT.72 Additional external staff and operational expenditure are financed through the Digital Europe Programme, mentioned above.73 It is undeniable that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s complex tasks come with a need for appropriate financing and staffing.74

17The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s organisational structure originally comprised five units:75 the Excellence in AI and Robotics Unit76, the Regulation and Compliance Unit77, the Unit on AI Safety78, the AI Innovation and Policy Coordination Unit79, and the AI for Societal Good Unit.80 In 2025, a sixth unit was added – the Unit for AI in Health and Life Sciences.81 Two advisors are also foreseen for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. :82 the Lead Scientific Advisor83 and the Advisor for International Affairs84. Upon unveiling the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. in May 2024, the Commission announced that it would eventually employ more than 140 staff.85 As of January 2026, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. employs more than 125 staff.86

18With regard to GPAI governance, the Unit on AI Safety and the Regulation and Compliance Unit are particularly relevant, as they are both tasked with ‘enforcing general-purpose AI rules’.87 Additionally, the Unit on AI Safety ‘focuses on the identification of systemic risks Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. , possible mitigation measures, as well as evaluation and testing approaches’.88

19The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. lacks legal personality.89 This is in line with its status as a Commission directorate.90 It is typical for acts establishing EU entities with legal personality to set this out expressly91 or, in the case of executive agencies established by the Commission, to reference Council Regulation (EC) No 58/2003 which establishes that such executive agencies have legal personality.92 The lack of such an express provision – such as the one previously contained in the European Parliament’s proposal that did not survive the trilogue negotiations93 – therefore indicates that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. does not have a separate legal personality.94 The mere fact that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has been afforded tasks under the Establishment Decision and the AI Act does not compel a different conclusion.95

20For questions of judicial review, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s lack of legal personality implies that the Commission will be the appropriate defendant in legal proceedings and will act on behalf of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .96 This is reinforced by the observation that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , despite its designation as an office, is not an office of the Union in the sense of Articles 263, 265 and 267 TFEU,97 as discussed in Section 3.4. below.

3.3. A special institutional status?

21Having established that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s nature is a Commission directorate without legal personality,98 the question arises of whether the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has a special institutional status among the Commission’s directorates, and in particular whether it enjoys some form of ‘operational autonomy’99 or independence from the Commission in the exercise of its functions. Several features distinguish the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. from other Commission directorates: its designation as the ‘ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’, its establishment by Commission decision,100 the fact that – unlike other directorates in the respective regulations101 – it is directly addressed in the AI Act, and the fact that the European Parliament’s proposal – which was ultimately not adopted – envisaged it as an independent body of the Union with legal personality.102

22Nonetheless, there are convincing arguments against the proposition that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has a special institutional status within the Commission.103 The Establishment Decision contains no indication in this direction, stating that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is established within the Commission and ‘should operate in accordance with Commission internal processes’.104 Importantly, in this regard, language indicating that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. can independently exercise Commission powers is absent from the Establishment Decision.105 Moreover, the AI Act’s legislative history, including a proposal that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. be an independent body of the Union with legal personality, does not determine the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s current legal status pursuant to the AI Act as enacted. It has rightly been pointed out that, under the AI Act now in force, ‘the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is a very different body than once proposed in the draft of the Parliament’.106

23Finally, the use of the term ‘ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ in the AI Act does not necessarily support its special institutional status. In fact, the AI Act’s use of this term, in many instances, fairly reflects the institutional status of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as a Commission directorate.107 Importantly, Article 3(47)’s definition of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , expressly referring to the Establishment Decision, provides that all references in the AI Act to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ‘shall be construed as references to the Commission’, providing a clear indication of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s subsumed status within the Commission. Article 88(1) further sets out that it is the Commission that is assigned the ‘power to supervise and enforce Chapter V [of the AI Act on GPAI models]’ (emphasis added), whereas the ‘implementation of these tasks’ (emphasis added) shall be entrusted to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .108 The instances where the AI Act appears, under a literal reading, to confer powers directly to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. provide only very limited support for a special institutional status,109 as these provisions do not directly concern the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s institutional nature. Moreover, it has rightly been pointed out that the ‘confusing’ use of the terms ‘ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ and ‘Commission’ may well be ‘the result of the political negotiations and the late changes to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’.110

24Some scholars have noted that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. resembles EU interinstitutional services,111 citing its ‘focused scope’ concerning the implementation of the AI Act and its provision of support to various institutions, agencies, and bodies such as the European Parliament, the Council, the Central Bank, the European Data Protection Board, the European Investment Bank, and the Board.112 However, while the Establishment Decision states that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ‘shall establish the appropriate forms of cooperation with bodies, offices and agencies of the Union’,113 it does not make express mention of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s support of the institutions cited in the literature, save for the Board.114 Moreover, a limited remit and a degree of interinstitutional cooperation are characteristic of Commission directorates generally.115 These features therefore do not confer any special institutional status on the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .116

3.4. Not an office under Articles 263, 265 and 267 TFEU

25The Treaties, including their provisions on judicial review under Articles 263, 265117 and 267118 TFEU, contain numerous references to ‘bodies, offices and agencies of the Union’ (emphasis added).119 In particular, according to Article 263(1) TFEU, the Court of Justice of the European Union reviews the legality of acts of Union offices which are intended to produce legal effects vis-à-vis third parties.120 The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is not a Union office under this provision. As laid out above,121 the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. forms part of the Commission, which is an institution of the Union122 and which the Treaties treat as distinct from ‘bodies, offices and agencies of the Union’.123 The fact that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is referred to as an ‘Office’ does not determine its legal nature.124 Article 6 of the Establishment Decision125 further reinforces that the Commission did not conceive the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as a Union office in the sense used by Articles 263, 265 and 267 TFEU.

3.5. Not a European office under Article 2(27) Regulation (EU, Euratom) 2024/2509

26The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is not a European office within the meaning of Article 2(27) of the Financial Regulation,126 as a footnote to Recital 7 of the Establishment Decision clarifies.127 Consequently, Articles 64 to 67 of Section 1., entitled ‘European offices’, of Chapter 3 of the Financial Regulation, containing provisions on the scope of competences, appropriations, tasks and accounting records of European offices, are inapplicable to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .

4. Remit

27There is no single provision that clearly establishes the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s remit.128 Rather, its remit emerges from reading the provisions of the AI Act and the Establishment Decision together,129 in particular Articles 3(47), 64 and 88(1) AI Act, and Articles 2 and 3 of the Establishment Decision.130 The tasks of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. concern the AI Act in particular131 but are not limited to it.132

28Given the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s wide range of tasks,133 legal scholarship has tried to identify the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s ‘primary mission’.134 Some argue that it is to ensure the AI Act’s harmonised implementation and enforcement,135 while others contend that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , at its core, serves as a liaison between the Commission and the governance structure established by the AI Act.136 A further view holds that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s core function is to support the Commission with regard to particularly complex tasks.137 However, while prioritisation will likely be key in practice, neither the AI Act nor the Establishment Decision expressly prioritises amongst the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s various tasks. Although the Establishment Decision lists the AI Act-related tasks before the additional tasks, and amongst the AI Act-related tasks places those concerning GPAI models first,138 the normative value of this appears limited. Moreover, Recital 148 only states rather broadly: ‘The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. was established by Commission Decision and has as its mission to develop Union expertise and capabilities in the field of AI and to contribute to the implementation of Union law on AI.’139 The weighting of the individual tasks within the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s broad mission therefore appears not to be prescribed in detail by law.

4.1. Implementation of the AI Act

29The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. performs various tasks concerning the implementation of the AI Act.140 To understand the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s remit in this regard, it is essential to consider the allocation of responsibilities within the AI Act’s governance framework.141 The AI Act establishes a division of oversight whereby, in principle, Member States’ market surveillance authorities Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. oversee compliance with the AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. obligations imposed by the AI Act,142 while compliance with the AI Act’s GPAI model provisions143 is overseen at the Union level.144 Article 75(1) AI Act modifies this allocation of responsibilities by providing that where an AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. is based on a GPAI model and both the model and the system are developed by the same provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. , compliance with the AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. obligations is overseen at the Union level rather than at the Member State level.145 Beyond this division of responsibilities and consistent with the aim of harmonisation, the AI Act provides for Union-level instruments, including the adoption of delegated146 and implementing147 acts, the issuance of Commission guidelines,148 and the development of codes of practice149 and codes of conduct150, which may concern obligations overseen at the Member State and Union levels.

30In this context, Articles 3(1) and (2) of the Establishment Decision set out a multitude of tasks that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall perform ‘for the purposes of implementing and enforcing [the AI Act]’.151 The wording of these paragraphs suggests a distinction between these provisions: Article 3(1) of the Establishment Decision contains ‘tasks stemming from the forthcoming Regulation’ (emphasis added), while Article 3(2) of the Establishment Decision contains tasks that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ‘in order to contribute to the effective implementation of the forthcoming Regulation, […] shall be tasked with’ (emphasis added). This formulation indicates that Article 3(1) of the Establishment Decision contains tasks with which the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is directly charged by the AI Act, whereas Article 3(2) of the Establishment Decision contains tasks that relate to the AI Act for which the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. cannot act on its own initiative but only upon assignment by the competent entity.152 However, closer examination, especially of non-English language versions of the AI Act, casts doubt on this distinction.153 Absent other convincing explanations for the existence of two separate lists of tasks,154 the interpretation of a specific task is likely influenced little, if at all, by whether it appears in the first or second paragraph of Article 3 of the Establishment Decision.

4.1.1. Tasks under Article 3(1) of the Establishment Decision

31Article 3(1) of the Establishment Decision contains a list of six tasks. Of particular relevance to GPAI model regulation are the tasks concerning the oversight of GPAI model compliance listed in Article 3(1)(a)–(d) of the Establishment Decision,155 namely:

a. ‘developing tools, methodologies and benchmarks for evaluating capabilities of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. , in particular for very large general purpose AI models with systemic risks Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. ’;156

32This task relates to the systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. classification of GPAI models under the AI Act. In particular, Article 51(1)(a) refers to the ‘technical tools and methodologies, including indicators and benchmarks’ for evaluating a model’s high-impact capabilities Article 3(64) AI Act: ‘high-impact capabilities’ means capabilities that match or exceed the capabilities recorded in the most advanced general-purpose AI models. ,157 while Annex XIII contains further criteria for determining whether a GPAI model meets the classification condition under Article 51(1)(b).158 Both Articles 51(3) and 52(4) empower the Commission to adopt delegated acts concerning these matters.159 The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s development of tools, methodologies and benchmarks under Article 3(1)(a) of the Establishment Decision will likely form the basis for such delegated acts.

33Additionally, Article 92 provides for the Commission’s power to conduct evaluations of a GPAI model both to assess its provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. ’s compliance with its obligations under the AI Act and ‘to investigate systemic risks Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. at Union level of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. with systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. ’.160 The task of developing evaluation tools, methodologies and benchmarks under Article 3(1)(a) of the Establishment Decision is likely to contribute to this function as well.

34As to the reference to ‘very large general purpose AI models with systemic risks Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. ’ in Article 3(1)(a) of the Establishment Decision, it should be noted that the AI Act does not itself make use of the term ‘very large’ in the context of GPAI models. This reference most likely denotes GPAI models that have been trained with a large amount of training compute above 1025 floating-point operations Article 3(67) AI Act: ‘floating-point operation’ means any mathematical operation or assignment involving floating-point numbers, which are a subset of the real numbers typically represented on computers by an integer of fixed precision scaled by an integer exponent of a fixed base. and are therefore automatically classified as GPAI models with systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. under Article 51(1)(a) and (2).161

b. ‘monitoring the implementation and application of rules on general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and systems, in particular where the model and the system are developed by the same provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. ’;162

35This task relates to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s monitoring activities concerning AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. based on a GPAI model where both the model and the system are developed by the same provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. under Article 75(1).163 It also relates to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s monitoring activities under Article 89 concerning GPAI models.164 With regard to the latter, Recital 164 provides that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ‘should be able to take the necessary actions to monitor the effective implementation of and compliance with the obligations for providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. laid down in this Regulation’.165

c. ‘monitoring the emergence of unforeseen risks Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. stemming from general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. , including by responding to alerts from the scientific panel;’166

36Alerts from the scientific panel to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. referred to in Article 3(1)(c) of the Establishment Decision are governed by Article 90 of the AI Act. According to this provision, ‘[t]he scientific panel may provide a qualified alert to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. where it has reason to suspect that: (a) a general-purpose AI model Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. poses concrete identifiable risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. at Union level; or (b) a general-purpose AI model Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. meets the conditions [for systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. classification of GPAI models] referred to in Article 51.’167 The provision further stipulates that ‘[u]pon such qualified alert, the Commission, through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. and after having informed the Board, may exercise the powers laid down in [Section 5. of Chapter IX of the AI Act] for the purpose of assessing the matter.’168

37The wording of Article 3(2)(c) of the Establishment Decision (‘including’) suggests that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is tasked with monitoring unforeseen risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. emergence from GPAI models not only in response to such alerts from the scientific panel but also on its own initiative. Such an interpretation aligns with the AI Act’s broader regulatory framework: Article 89 governs the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s monitoring activities in relation to GPAI model providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. without expressly referencing the scientific panel, and Articles 51(1)(b) and 52(4) provide, in the context of the classification of GPAI models as presenting systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. , that the Commission may designate GPAI models as presenting systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. either following a qualified alert from the scientific panel or ‘ex officio’.169

d. ‘investigating possible infringements of rules on general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and systems, including by collecting complaints and alerts, assisting in the preparation of decisions of the Commission and conducting evaluations pursuant to the forthcoming Regulation’.170

38This task relates to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role in the oversight of GPAI models under Section 5. (‘Supervision, investigation, enforcement and monitoring in respect of providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. ’) of Chapter IX of the AI Act.171 It encompasses various provisions in that Section. First, it concerns Article 88(1), according to which the Commission ‘shall entrust’ the implementation of its supervision and enforcement tasks concerning GPAI model obligations under Chapter V of the AI Act to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .172 The reference to ‘collecting complaints and alerts’ concerns complaints by downstream providers Article 3(68) AI Act: ‘downstream provider’ means a provider of an AI system, including a general-purpose AI system, which integrates an AI model, regardless of whether the AI model is provided by themselves and vertically integrated or provided by another entity based on contractual relations. governed by Article 89(2) and alerts by the scientific panel under Article 90. The phrase ‘assisting in the preparation of decisions of the Commission’ may relate, in particular, to Commission requests for documentation and information under Article 91(1), requests for access to GPAI models for the purposes of conducting evaluations under Article 92(3), and requests for measures under Article 93(1).173 Finally, the reference to ‘conducting evaluations’ concerns the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s evaluations of GPAI models under Article 92(1).

39It is less clear why Article 3(2)(d) of the Establishment Decision refers to GPAI systems. The AI Act defines a GPAI system in Article 3(66) as ‘an AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. which is based on a general-purpose AI model Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and which has the capability to serve a variety of purposes, both for direct use as well as for integration in other AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. ’. While the AI Act references GPAI systems in various instances,174 it does so without establishing obligations exclusive to GPAI systems. Notably, competence for oversight over GPAI systems lies – since they are AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. – primarily with the Member States rather than the Commission.175 The Commission monitors and supervises compliance only for GPAI systems developed by the provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of the GPAI model they are based on, as provided in Article 75(1). Against this background, Article 3(2)(d) of the Establishment Decision may not refer to GPAI systems in general but rather to GPAI systems developed by the provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of the GPAI model they are based on. On this reading, Article 3(2)(d) of the Establishment Decision would – just as Article 3(2)(b) of the Establishment Decision does – refer to Article 75(1), under which the Commission monitors and supervises compliance of such systems with AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. obligations under the AI Act.

4.1.2. Tasks under Article 3(2) of the Establishment Decision

40Beyond the tasks laid out in Article 3(1) of the Establishment Decision, Article 3(2) of the Establishment Decision tasks the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. with further responsibilities ‘in order to contribute to the effective implementation of the [AI Act]’. These responsibilities encompass nine additional tasks. In particular, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is tasked with ‘assisting the Commission’176 in the preparation of (i) Commission decisions, implementing and delegated acts;177 (ii) guidance and guidelines aimed at supporting the AI Act’s practical implementation;178 and (iii) standardisation requests.179

41As regards the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. supporting the Commission’s issuance of guidance and guidelines, it is important to emphasise that such guidance is not limited to guidelines on the list of topics provided in Article 96(1).180 As evidenced by the provision’s wording (‘in particular’), this list is non-exhaustive.181 Moreover, while Article 96(1) references ‘guidelines’, Commission guidance aimed at supporting the practical implementation of Union law may take a variety of forms, as reflected in the Commission’s established practice in relation to other Union legislation.182 Such guidance has, for example, been issued by way of notices,183 templates,184 guidance letters,185 comfort letters,186 and Q&As.187 The Commission’s role of taking appropriate initiatives to promote the Union’s general interest and ensuring and overseeing the application of Union law under Article 17(1) TEU supports the issuance of such guidance.188

42Moreover, Article 3(2) of the Establishment Decision tasks the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. with ‘facilitating the [AI Act’s] uniform application’.189 Concerning Union-level governance, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall ‘coordinat[e] the establishment of an effective governance system’,190 ‘provid[e] the Secretariat for the AI Board and its subgroups’191 and ‘provid[e] administrative support to the advisory forum and the scientific panel, where applicable’.192 The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s responsibilities further include contributions to the drawing up of codes of practices and codes of conduct,193 as well as the establishment and operation of AI regulatory sandboxes Article 3(55) AI Act: ‘AI regulatory sandbox’ means a controlled framework set up by a competent authority which offers providers or prospective providers of AI systems the possibility to develop, train, validate and test, where appropriate in real-world conditions, an innovative AI system, pursuant to a sandbox plan for a limited time under regulatory supervision. 194 and the conduct of AI Act-related evaluations, reviews and reports.195

4.1.3. Internal allocation of responsibilities within the Commission

43Regarding the internal allocation of responsibilities within the Commission between the Commission as the College of Commissioners and the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as part of the Commission’s permanent bureaucracy,196 it appears that in principle the power to issue legally binding acts, such as delegated acts,197 implementing acts,198 designation decisions,199 requests vis-à-vis GPAI model providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. ,200 or the imposition of fines,201 remains with the Commission, while the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. supports the Commission in exercising these powers by preparing draft decisions and, once approved, communicating them externally on the Commission’s behalf.202 This allocation of responsibilities, which accords with the nature of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as a Commission directorate203 without separate legal personality204 and the principle of collegiate responsibility governing the adoption of Commissions actions205, is reflected in numerous provisions in the AI Act206 and the Establishment Decision207 and finds further support in the Commission Guidelines.208 In particular, according to Article 88(1), it is the Commission that is assigned the ‘power to supervise and enforce Chapter V [of the AI Act on GPAI models]’ (emphasis added), whereas the ‘implementation of these tasks’ (emphasis added) shall be entrusted to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .209

44This interpretation does not appear fundamentally challenged by those provisions under the AI Act210 which appear, by their wording, to assign powers directly to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .211 Notably, Article 75(1) states that ‘the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall have powers to monitor and supervise compliance’ of AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. based on GPAI models that are developed by the GPAI model’s provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. and that ‘[t]o carry out its monitoring and supervision tasks, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall have […] powers of a market surveillance authority Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. ’ (emphasis added).212 Were the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. able to exercise powers under this provision independently of the Commission as a collegiate body, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. would occupy a unique position within the Commission,213 comparable only, to a limited extent, to that of OLAF.214 However, it appears doubtful that the legislature intended to confer such a unique status upon the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .215 It must be recalled that, pursuant to Article 3(47), all references to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. must be construed as references to the Commission. This implies that the AI Act does not confer any exclusive powers on the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. but rather, in line with the principle of collegiate responsibility, empowers the Commission as a collegiate body.216 Separate supervision and enforcement powers of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. would therefore be necessarily duplicative and, at least in the context of Article 75(1), likely contradict the provision’s purpose ‘to avoid overlapping competences’.217 Moreover, unlike in the case of OLAF,218 the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s Establishment Decision contains no indication that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. may independently exercise the Commission’s powers.219 Concerning the obligations for GPAI model providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. specifically, Article 88(1) clarifies that it is the Commission that is empowered to supervise and enforce Chapter V, whereas the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. implements these tasks.220

4.2. Further tasks and international cooperation

45Unlike the AI Act-related tasks outlined in Articles 2(1) and 3, Article 2(2) of the Establishment Decision assigns the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. four ‘additional tasks’ with a broader remit.221 These encompass222 (i) contributing to international AI cooperation,223 (ii) supporting actions and policies that harness AI’s societal and industrial benefits,224 (iii) promoting the development and adoption of beneficial AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. and applications,225 and (iv) monitoring AI markets and technologies.226

46The Establishment Decision envisages that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. supports the Commission’s international cooperation by promoting the ‘responsible stewardship of AI’ and the Union’s ‘approach to trustworthy’ AI.227 The Establishment Decision also foresees the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. contributing to international AI regulation and governance cooperation as well as the implementation of international AI agreements.228 As part of this international effort, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. forms part of the International Network of AI Safety Institutes, launched in November 2024.229 Consisting of AI safety and security institutions from Australia, Canada, the European Commission, France, Japan, Kenya, the Republic of Korea, Singapore, the United Kingdom, and the United States, the network aims ‘to catalyze a new phase of international cooperation on AI safety’ by advancing research on AI risks Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. and capabilities, developing best practices for testing, and sharing relevant information and technical tools internationally.230 In this context, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. contributed to the third joint testing exercise of the International Network of AI Safety Institutes in July 2025 concerning the evaluation of AI agents.231

5. Operative functioning

47According to Recital 7 of the Establishment Decision, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ‘should operate in accordance with Commission internal processes’.232 As the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is a Commission directorate,233 the Commission’s rules of procedure apply.234

48In addition, Articles 3(3) and 4–7 of the Establishment Decision govern how the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. should perform its various tasks,235 requiring it to cooperate with (i) stakeholders;236 (ii) other Commission Directorate-Generals and services;237 (iii) Union bodies, offices and agencies;238 (iv) Member States authorities and bodies;239 and (v) international actors240. Unsurprisingly though, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s functions are ‘without prejudice to the functions of other Commission departments in their respective areas of responsibility’.241

49Article 78 AI Act provides that confidentiality of information and data obtained by inter alia the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. in carrying out its tasks is to be maintained.242 The importance of such confidentiality is underscored by Recital 161 AI Act.243

6. Evaluation and review under Article 112(5)

50The Commission must evaluate the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s functioning by 2 August 2028.244 As part of this evaluation, the Commission must address two questions expressly laid out in Article 112(5): ‘whether the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has been given sufficient powers and competences to fulfill its tasks’ and ‘whether it would be relevant and needed for the proper implementation and enforcement of this Regulation to upgrade the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. and its enforcement competences and to increase its resources.’245 For this purpose, the Commission must take into account the positions and findings of relevant bodies and sources, such as the Board, the European Parliament and the Council.246 It may also request information from the Board, the Member States and national competent authorities Article 3(48) AI Act: ‘national competent authority’ means a notifying authority or a market surveillance authority; as regards AI systems put into service or used by Union institutions, agencies, offices and bodies, references to national competent authorities or market surveillance authorities in this Regulation shall be construed as references to the European Data Protection Supervisor. .247 If necessary, particularly, for present purposes, for any upgrades or increases of competences or resources for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , the Commission shall submit proposals to amend the AI Act.248

51Moreover, and relatedly, the Commission must carry out an assessment of the AI Act’s enforcement and submit a report on this matter to the European Parliament, the Council and the European Economic and Social Committee by 2 August 2031.249 The assessment may also include the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role and functioning, as the report ‘shall, where appropriate, be accompanied by a proposal for amendment of this Regulation with regard to the structure of enforcement and the need for a Union agency to resolve any identified shortcomings.’250

  1. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) [2024] OJ L 1689/1 (“AI Act”). ↩︎
  2. See Section 2. ↩︎
  3. See AI Act, recital 96, last sentence. Further, see Commission Decision of 24 January 2024 establishing the European Artificial Intelligence Office (C/2024/1459) (“Establishment Decision”), art 1(1). ↩︎
  4. See Establishment Decision (n 3) art 1(2): ‘The Office shall be part of the administrative structure of the Directorate-General for Communication Networks, Content and Technology (the “Directorate-General”).’; see European Union, ‘Artificial Intelligence Office (CNECT.A)’ (EU Whoiswho) <https://op.europa.eu/en/web/who-is-who/organization/-/organization/COM/COM_CRF_3785> accessed 4 January 2026; Section 3.1. ↩︎
  5. For the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role in EU AI governance, see, in particular, Establishment Decision (n 3) arts 2 and 3; AI Act, arts 3(47) and 88(1); Eric Hilgendorf and Johannes Härtlein, ‘Art. 64 Büro für Künstliche Intelligenz’ in Eric Hilgendorf and Johannes Härtlein (eds), KI-VO: Verordnung über Künstliche Intelligenz: Kommentar (Nomos 2025) para 1; Claudio Novelli and others, ‘A Robust Governance for the AI Act: AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , AI Board, Scientific Panel, and National Authorities’ (2025) 16 European Journal of Risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. Regulation 566, 575; for an overview over the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s remit, see Section 4. ↩︎
  6. David Roth-Isigkeit, ‘Art. 64 Büro für Künstliche Intelligenz’ in Mario Martini and Christiane Wendehorst (eds), KI-VO: Verordnung über Künstliche Intelligenz: Kommentar (2nd edn, C H Beck 2026) para 23; Novelli and others (n 5) 575; Björn Herbers and David Rappenglück, ‘Die Durchsetzung der KI-Verordnung auf EU-Ebene: Das EU AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ (2024) Recht Digital 432, 433, para 4; see AI Act, recital 164; see Section 4.1.1. and Section 4.1.2. ↩︎
  7. Roth-Isigkeit (n 6) para 3. For the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s remit within the Commission, see Section 4.1.3. For its role in the supervision and enforcement of GPAI model obligations under the AI Act, see commentary on Article 88, Section 2.1.3. ↩︎
  8. See Anne Paschke and Sarah Rachut, ‘Art. 64 Büro für Künstliche Intelligenz’ in Jens Schefzig and Robert Kilian (eds), Beck’scher Online-Kommentar KI-Recht (4th edn, C H Beck 2025) para 3; Marieke Merkle, ‘Art. 3 Begriffsbestimmungen’ in David Bomhard, Fritz-Ulli Pieper and Susanne Wende (eds), Kommentar KI-VO: Verordnung über Künstliche Intelligenz (Fachmedien Recht und Wirtschaft 2025) para 359. The Commission has provided details on the establishment of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. in a press release (see European Commission, ‘Commission Establishes AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. to Strengthen EU Leadership in Safe and Trustworthy Artificial Intelligence’ (Press Release IP/24/2982, May 29, 2024) <https://ec.europa.eu/commission/presscorner/detail/en/ip_24_2982> (“ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release”). ↩︎
  9. See Establishment Decision (n 3) recitals 2 and 8. ↩︎
  10. Establishment Decision (n 3) recital 8. ↩︎
  11. A directorate is an administrative subdivision of a Directorate-General, typically containing several units (see Commission Decision (EU) 2024/3080 of 4 December 2024 establishing the Rules of Procedure of the Commission and amending Decision C(2000) 3614 (“Commission Rules of Procedure”), art 45(1), third sentence). For the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s nature as a Commission directorate, see Section 3.1. ↩︎
  12. European Parliament, ‘Amendments adopted on 14 June 2023 on the proposal for a regulation of the European Parliament and of the Council on laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts’ P9_TA(2023)0236, (“Parliament Amendments”) Amendments 524–530; see Matthias Schmidl and Andreas Rohner, ‘Article 64 AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ in Ceyhun Necati Pehlivan, Nikolaus Forgó and Peggy Valcke (eds), The EU Artificial Intelligence (AI) Act: A Commentary (Wolters Kluwer 2024) 983, s 3.1; Jonathan Kirschke-Biller and Anna Lena Füllsack, ‘Art. 3 Begriffsbestimmungen’ in Jens Schefzig and Robert Kilian (eds), Beck’scher Online-Kommentar KI-Recht (4th edn, C H Beck 2025) para 546. ↩︎
  13. Clemens Bernsteiner and Thomas Rainer Schmitt, ‘Art. 88 Durchsetzung der Pflichten der Anbieter von KI-Modellen mit allgemeinem Verwendungszweck’ in Mario Martini and Christiane Wendehorst (eds), KI-VO: Verordnung über Künstliche Intelligenz: Kommentar (2nd edn, C H Beck 2026) para 10; Schmidl and Rohner (n 12) 983, s 3.1. ↩︎
  14. See European Commission, ‘Proposal for a Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act) and Amending Certain Union Legislative Acts’ COM (2021) 206 final. For an overview of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s legislative history concerning GPAI model supervision and enforcement, see commentary on Article 88, Section 1.2. in this work; for a general overview over the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s legislative history, see Susanne Wende, ‘Art. 64 Büro für Künstliche Intelligenz’ in David Bomhard, Fritz-Ulli Pieper and Susanne Wende (eds), Kommentar KI-VO: Verordnung über Künstliche Intelligenz (Fachmedien Recht und Wirtschaft 2025) paras 5–8; see also Schmidl and Rohner (n 12) 983, s 3.1. ↩︎
  15. Schmidl and Rohner (n 12) 983, s 3.1; for an overview over AI Act provisions with relevance to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , see Kirschke-Biller and Füllsack (n 12) paras 547–549. The AI Act’s approach of specifically addressing a Commission directorate – the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. – appears unusual, given that other EU regulations such as Regulation (EU) 2022/1925 of the European Parliament and of the Council of 14 September 2022 on contestable and fair markets in the digital sector and amending Directives (EU) 2019/1937 and (EU) 2020/1828 (Digital Markets Act) [2022] OJ L 265/1 (“DMA”) and Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 October 2022 on a Single Market For Digital Services and amending Directive 2000/31/EC (Digital Services Act) [2022] OJ L 277/1 (“DSA”) only refer to the Commission as a whole. It may be explained by the AI Act’s legislative history, in particular by the European Parliament’s proposal originally conceiving the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as an independent body of the Union with separate legal personality (see Parliament Amendments (n 12) Amendments 524–530). For a further discussion of whether the AI Act’s approach entails a special institutional status for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , see Section 3.3. ↩︎
  16. See Kirschke-Biller and Füllsack (n 12) paras 547–549 who emphasise the central role of article 64; further, see article 75(1) AI Act with regard to the monitoring and supervision of compliance of AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. based on a GPAI model, where the model and the system are developed by the same provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. . ↩︎
  17. Kirschke-Biller and Füllsack (n 12) para 543; see also Section 3.1. ↩︎
  18. See AI Act, ch VII, title. ↩︎
  19. For an analysis of article 64’s first paragraph, see Section 2.1..; for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s mission, see also AI Act, recital 148, third sentence (‘The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. was established by Commission Decision and has as its mission to develop Union expertise and capabilities in the field of AI and to contribute to the implementation of Union law on AI.’); Establishment Decision (n 3) recital 4 (‘It is necessary to develop expertise and capabilities at Union level with a view to foster such understanding, to contribute to the implementation and enforcement of the forthcoming Regulation laying down harmonised rules on artificial intelligence, and to contribute to the implementation of international rules and principles on AI, such as the G7 Code of Conduct and Guiding Principles for developers of advanced AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. .’); Establishment Decision (n 3) art 2. ↩︎
  20. For an analysis of article 64’s second paragraph, see Section 2.2. AI Act, recital 148, fourth sentence, further elaborates this mandate: ‘Member States should facilitate the tasks of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. with a view to support the development of Union expertise and capabilities at Union level and to strengthen the functioning of the digital single market.’ ↩︎
  21. See, AI Act, ch IX, s 5, title. ↩︎
  22. For an analysis of this provision, see commentary on Article 88, Section 2.1.4. in this work. ↩︎
  23. See, in particular, AI Act, arts 88(1), 89, 90(1) and (2), 91(2), 92(1) and 93(2) for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role in the monitoring, supervision and enforcement of GPAI model provisions and AI Act, recitals 148 and 161–164 for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role in the AI Act’s governance framework. For a list of AI Act provisions mentioning the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , see Kirschke-Biller and Füllsack (n 12) 548.1–548.3. ↩︎
  24. See Roth-Isigkeit (n 6) paras 2–8; Paschke and Rachut (n 8) para 4; Wende (n 14) para 2; Kirschke-Biller and Füllsack (n 12) para 545. For the legislature’s conception of the AI Act’s governance framework, see, in particular, AI Act, recital 148. ↩︎
  25. The Board, established by the AI Act (AI Act, art 65(1)), is composed of representatives from the Member States, with the European Data Protection Supervisor participating as an observer and the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. attending AI Board meetings without voting rights (AI Act, art 65(2)). The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. acts as the secretariat for the Board (AI Act, art 65(8)). ↩︎
  26. The advisory forum, whose establishment is provided for in article 67(1), shall be composed members representing a ‘balanced selection of stakeholders, including industry, start-ups, SMEs, civil society and academia’ (AI Act, art 67(2), first sentence; see also commentary on Article 67, Section 2.3.2. in this work). In July 2025, the Commission launched a call for applications to join the advisory forum (European Commission, ‘Call for Expression of Interest for Appointment of Members of the Advisory Forum Pursuant to Article 67 of the Artificial Intelligence Act’ (2025) https://ec.europa.eu/newsroom/dae/redirection/document/118257 accessed 14 January 2026) and published its terms of reference (European Commission, ‘Informal Commission Expert Group: AI Act Advisory Forum: Terms of Reference’ (2025) https://ec.europa.eu/newsroom/dae/redirection/document/118258 accessed 14 January 2026). ↩︎
  27. The scientific panel, established by article 2(1) of Commission Implementing Regulation (EU) 2025/454 of 7 March 2025 laying down the rules for the application of Regulation (EU) 2024/1689 of the European Parliament and of the Council as regards the establishment of a scientific panel of independent experts in the field of artificial intelligence [2025] OJ L 2025/454 on the basis of article 68(1), shall consist of experts with ‘particular expertise and competence and scientific or technical expertise in the field of AI’ (see AI Act, art 68(2)(a); see also commentary on Article 68, Section 2.3.2. in this work). In June 2025, the Commission launched a call for applications to join the scientific panel (Commission, ‘Call for Expression of Interest for the Establishment of a Scientific Panel of Independent Experts in the Field of Artificial Intelligence and its Impacts’ (2025) <https://ec.europa.eu/newsroom/dae/redirection/document/117015> accessed 14 January 2026). ↩︎
  28. See Hilgendorf and Härtlein (n 5) para 1. ↩︎
  29. See, in particular, Establishment Decision (n 3) art 3(2)(g) and (h); AI Act, recital 148, second and third sentence, art 65(8); see also Section 4.1.2., para 42. ↩︎
  30. See AI Act, art 66: ‘The Board shall advise and assist the Commission and the Member States in order to facilitate the consistent and effective application of this Regulation. To that end, the Board may in particular: […] (c) provide advice on the implementation of this Regulation, in particular as regards the enforcement of rules on general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. […].’; AI Act, art 67(1): ‘An advisory forum shall be established to provide technical expertise and advise the Board and the Commission, and to contribute to their tasks under this Regulation.’ (for an analysis of article 67, see commentary on Article 67 in this work); AI Act, art 68(3): ‘The scientific panel shall advise and support the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , in particular with regard to the following tasks: (a) supporting the implementation and enforcement of this Regulation as regards general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and systems, in particular by […].’ (for an analysis of this provision, see commentary on Article 68 in this work). ↩︎
  31. See Section 3.1. ↩︎
  32. See Section 3.2. ↩︎
  33. See Section 3.3. ↩︎
  34. See Section 3.4. ↩︎
  35. See Section 3.5. ↩︎
  36. See Section 4.1. ↩︎
  37. See Section 4.2. ↩︎
  38. See commentary on Article 88 in this work. ↩︎
  39. See Section 2.1. ↩︎
  40. See Section 2.2. ↩︎
  41. See Schmidl and Rohner (n 12) 983, s 3.1 (‘Article 64 itself consists of two rather abstract provisions that do not directly address the tasks of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. .’); Roth-Isigkeit (n 6) paras 9, 11. For a list of provisions concerning the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role concerning GPAI model governance, see Kirschke-Biller and Füllsack (n 12) para 548.1. For the establishment and nature of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , see Section 3.1. For its remit, see Section 4. ↩︎
  42. Schmidl and Rohner (n 12) 983, s 3.1; see Paschke and Rachut (n 8) para 12 who argue that article 64(1) lays out the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s foundational remit (‘grundlegenden Aufgabenbereich’). ↩︎
  43. See, in particular, AI Act, recital 148, second and third sentence: ‘The effective implementation and enforcement of this Regulation require a governance framework that allows to coordinate and build up central expertise at Union level. The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. was established by Commission Decision and has as its mission to develop Union expertise and capabilities in the field of AI and to contribute to the implementation of Union law on AI.’; Establishment Decision (n 3) recital 4: ‘It is necessary to develop expertise and capabilities at Union level with a view to foster such understanding, to contribute to the implementation and enforcement of the forthcoming Regulation laying down harmonised rules on artificial intelligence [the AI Act], and to contribute to the implementation of international rules and principles on AI, such as the G7 Code of Conduct and Guiding Principles for developers of advanced AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. .’; see also Paschke and Rachut (n 8) paras 12 and 13. ↩︎
  44. See Regulation (EU) 2021/694 of the European Parliament and of the Council of 29 April 2021 establishing the Digital Europe Programme and repealing Decision (EU) 2015/2240 [2021] OJ L 166/1 (“Digital Europe Programme Regulation”), arts 3(1) and 9(1); European Commission, ‘The Digital Europe Programme’ (2025) <https://digital-strategy.ec.europa.eu/en/activities/digital-programme> accessed 22 March 2026. ↩︎
  45. Digital Europe Programme Regulation (n 43) art 5(1)(a). ↩︎
  46. Establishment Decision (n 3) art 8(2) and (3); see also Section 3.1. ↩︎
  47. Paschke and Rachut (n 8) para 12 specify that ‘expertise’ encompasses knowledge about various aspects of AI, such as data processing, machine learning, fields of application and market developments. ↩︎
  48. Paschke and Rachut (n 8) para 12 describe ‘capabilities’ as concerning technical and practical competences regarding the development and evaluation of AI models. ↩︎
  49. AI Act, recital 148, sixth sentence’; see also Paschke and Rachut (n 8) para 15. ↩︎
  50. AI Act, recital 148, sixth sentence. ↩︎
  51. See Wende (n 14) para 3. ↩︎
  52. See AI Act, art 3(47) and art 88(1); for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s remit, see Section 4. ↩︎
  53. See Paschke and Rachut (n 8) para 23; Roth-Isigkeit (n 6) para 17; see also AI Act, recital 148, fourth sentence: ‘Member States should facilitate the tasks of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. with a view to support the development of Union expertise and capabilities at Union level and to strengthen the functioning of the digital single market.’ For a discussion of the application of article 64(2) in the context of the enforcement of GPAI model obligations, see commentary on Article 88, Section 2.1.1. in this work. ↩︎
  54. See Roth-Isigkeit (n 6) para 17 who argues that article 64(2) does not extend beyond this duty of cooperation; see also Wende (n 14) paras 3, 30. According to the duty of sincere cooperation, ‘the European Union and the Member States are, in full mutual respect, to assist each other in carrying out the tasks provided for by the Treaties.’ (Case C-620/16, European Commission v Federal Republic of Germany [2019] ECLI:EU:C:2019:256, para 92). ↩︎
  55. Consolidated versions of the Treaty on European Union and the Treaty on the Functioning of the European Union [2012] OJ C 326/47 (“TEU”, “TFEU”). ↩︎
  56. This resemblance provides a counterargument against the view advanced by Wende (n 14) para 30 that article 64(2)’s duty goes beyond the Member States’ duty of cooperation under the Treaties. ↩︎
  57. See, for example, AI Act, art 75(2): ‘Where the relevant market surveillance authorities Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. have sufficient reason to consider general-purpose AI systems Article 3(66) AI Act: ‘general-purpose AI system’ means an AI system which is based on a general-purpose AI model and which has the capability to serve a variety of purposes, both for direct use as well as for integration in other AI systems. that can be used directly by deployers Article 3(4) AI Act: ‘deployer’ means a natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity. for at least one purpose that is classified as high-risk pursuant to this Regulation to be non-compliant with the requirements laid down in this Regulation, they shall cooperate with the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. to carry out compliance evaluations, and shall inform the Board and other market surveillance authorities Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. accordingly.’ (emphasis added). ↩︎
  58. See, for example, Cases C‑154/21 RW v Österreichische Post AG [2023] ECLI:EU:C:2023:3, para 29 and C-31/17 Cristal Union, the legal successor to Sucrerie de Toury SA v Ministre de l’Économie et des Finances [2018] ECLI:EU:C:2018:168, para 41; see also Koen Lenaerts and José A. Gutiérrez-Fons, ‘To Say What the Law of the EU Is: Methods of Interpretation and the European Court of Justice’ (2014) 20.2 Columbia Journal of European Law 3, 17–21. ↩︎
  59. For these tasks, see Establishment Decision (n 3) arts 2(2) and 4–7; Section 4.2. ↩︎
  60. See, for example, European Commission v Federal Republic of Germany (n 54) paras 92–100 concerning the infringement of a Member States’ duty of cooperation under article 4(3) TEU. ↩︎
  61. See Establishment Decision (n 3) art 2(3): ‘In the performance of the tasks referred to in paragraphs 1 and 2 of this Article, the Office shall: […] (d) cooperate with authorities and bodies of the Member States on behalf of the Commission.’; see also Roth-Isigkeit (n 6) para 17 who argues that point (d) of article 2(3) of the Establishment Decision confirms the Member States’ duty of cooperation. ↩︎
  62. See Establishment Decision (n 3) art 1(2), recital 6; Wende (n 14) paras 4, 15; Hilgendorf and Härtlein (n 5) para 3; Paschke and Rachut (n 8) para 7; Schmidl and Rohner (n 12) 983, s 3.1; Bernsteiner and Schmitt (n 13) para 14; Herbers and Rappenglück (n 6) para 8; Merkle (n 8) paras 362 and 363 who highlights that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. therefore is not politically independent; Hans Graux and others, ‘Interplay Between the AI Act and the EU Digital Legislative Framework’ (Study for the European Parliament’s Committee on Industry, Research and Energy, PE 778.575, 2025) <https://www.europarl.europa.eu/RegData/etudes/STUD/2025/778575/ECTI_STU(2025)778575_EN.pdf> accessed 9 January 2025, 70–71; Kasia Söderlund and Stefan Larsson, ‘Enforcement Design Patterns in EU Law: An Analysis of the AI Act’ Digital Society (2024) 3:41 <https://doi.org/10.1007/s44206-024-00129-8> accessed 9 January 2025, 12. For the role of Directorates-General within the Commission, see Commission Rules of Procedure (n 11) art 45(1); Koen Lenaerts, Piet Van Nuffel and Tim Corthaut, EU Constitutional Law (Oxford University Press 2021) para 12.071. ↩︎
  63. A directorate is an administrative division of a Directorate-General (see Commission Rules of Procedure (n 11) art 45(1), third sentence). ↩︎
  64. Commission Rules of Procedure (n 11) art 45(2). ↩︎
  65. Schmidl and Rohner (n 12) 983, s 3.1. ↩︎
  66. European Commission, ‘Communication from the Commission – Commission Guidelines on the Scope of the Obligations for Providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of General-Purpose AI Models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. Established by Regulation (EU) 2024/1689 (AI Act)’ C(2025) 7719 final (“Commission Guidelines”) para 90. ↩︎
  67. Establishment Decision (n 3) recital 6; see Paschke and Rachut (n 8) para 7. ↩︎
  68. Establishment Decision (n 3) recital 7. While this recital does not clarify the exact meaning of ‘Commission internal processes’, it implies that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is not intended to operate on the basis of special internal processes (see Section 3.2.). ↩︎
  69. Bernsteiner and Schmitt (n 13) para 15. ↩︎
  70. See Wende (n 14) para 15; Bernsteiner and Schmitt (n 13) para 15 (deriving this from the principle of collegiate responsibility). ↩︎
  71. See Wende (n 14) para 15. ↩︎
  72. Establishment Decision (n 3) art 8(1); see Roth-Isigkeit (n 6) para 13; Paschke and Rachut (n 8) para 8; see also AI Act, recital 148, sixth sentence: ‘The development of Union expertise and capabilities should also include making use of existing resources and expertise […].’ ↩︎
  73. Establishment Decision (n 3) art 8(2) and (3); see Roth-Isigkeit (n 6) para 13; Paschke and Rachut (n 8) para 9. For the Digital Europe Programme, see Section 2.1. ↩︎
  74. See Oskar J. Gstrein, Noman Haleem and Andrej Zwitter, ‘General-purpose AI regulation and the European Union AI Act’ (2024) 13(3) Internet Policy Review https://doi.org/10.14763/2024.3.1790 accessed 9 January 2025, s 5; see also Wende (n 14) para 16. For the tasks of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. see Section 4. ↩︎
  75. AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8); see also Wende (n 14) para 12; Hilgendorf and Härtlein (n 5) para 3; Herbers and Rappenglück (n 6) para 12. ↩︎
  76. The Excellence in AI and Robotics Unit ‘supports and funds research and development to foster an ecosystem of excellence’ and ‘coordinates the GenAI4EU initiative, stimulating the development of models and their integration into innovative applications’, according to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  77. The Regulation and Compliance Unit ‘coordinates the regulatory approach to facilitate the uniform application and enforcement of the AI Act across the Union, working closely with Member States’ and ‘will contribute to investigations on possible infringements, administering sanctions’, according to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  78. The Unit on AI Safety ‘focus[es] on the identification of systemic risks Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. of very capable general-purpose models, possible mitigation measures as well as evaluation and testing approaches’, according to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  79. The AI Innovation and Policy Coordination Unit ‘oversees the execution of the EU AI strategy, monitoring trends and investment, stimulating the uptake of AI through a network of European Digital Innovation Hubs and the establishment of AI Factories, and fostering an innovative ecosystem by supporting regulatory sandboxes and real-world testing’, according to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  80. The AI for Societal Good Unit ‘design[s] and implement[s] the international engagement of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. in AI for good, such as weather modelling, cancer diagnoses and digital twins for reconstruction’, according to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  81. See European Union, ‘Artificial Intelligence Office (CNECT.A)’ (n 4). ↩︎
  82. AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8); European Commission, ‘European AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. https://digital-strategy.ec.europa.eu/en/policies/ai-office accessed 5 January 2026. ↩︎
  83. The Lead Scientific Advisor’s role is ‘to ensure scientific excellence in evaluation of models and innovative approaches’, according to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  84. The Advisor for International Affairs’s role is ‘to follow up on [the Commission’s] commitment to work closely with international partners on trustworthy AI’, according to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  85. AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8). ↩︎
  86. European Commission, ‘European AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ <https://digital-strategy.ec.europa.eu/en/policies/ai-office> accessed 5 January 2026. ↩︎
  87. AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. Establishment Press Release (n 8); European Commission, ‘Interest in the European AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. https://ec.europa.eu/eusurvey/runner/AIOffice-Interest-General accessed 26 October 2025. ↩︎
  88. ibid. ↩︎
  89. Graux and others (n 62) 10; Novelli and others (n 5) 577; Bernsteiner and Schmitt (n 13) para 14; Wende (n 14) para 15; Herbers and Rappenglück (n 6) para 8; Jasper Siems, ‘Art. 88 Durchsetzung der Pflichten der Anbieter von KI-Modellen mit allgemeinem Verwendungszweck’ in David Bomhard, Fritz-Ulli Pieper and Susanne Wende (eds), Kommentar KI-VO: Verordnung über Künstliche Intelligenz (Fachmedien Recht und Wirtschaft 2025) para 19 (comparing the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. to the European Centre for Algorithmic Transparency which forms part of the Commission’s Joint Research Centre). ↩︎
  90. See Bernsteiner and Schmitt (n 13) para 14; compare, Case T-690/13 In vivo OOO v European Commission [2015] ECLI:EU:T:2015:519, para 14 with regard to European Anti-Fraud Office (“OLAF”): ‘It should be noted that OLAF, established by Commission Decision 1999/352/EC, ECSC, Euratom of 28 April 1999 (OJ 1999 L 136, p. 20), is an internal department of the Commission, the independence of which is purely functional and limited to its investigation activities. Accordingly, and in the absence of any contrary provision, OLAF does not have separate legal personality and it is the Commission which acts on its behalf in legal proceedings.’ ↩︎
  91. See Council Regulation (EU) 2017/1939 of 12 October 2017 implementing enhanced cooperation on the establishment of the European Public Prosecutor’s Office (‘the EPPO’) [2017] OJ L 283/1, art 3(2): ‘The EPPO shall have legal personality.’; Regulation (EC) No 726/2004 of the European Parliament and of the Council of 31 March 2004 laying down Union procedures for the authorisation and supervision of medicinal products for human use and establishing a European Medicines Agency [2004] OJ L 136/1, art 71: ‘The Agency shall have legal personality. In all Member States it shall enjoy the most extensive legal capacity accorded to legal persons under their laws. It may in particular acquire or dispose of movable and immovable property and may be a party to legal proceedings.’; Regulation (EC) No 1406/2002 of the European Parliament and of the Council of 27 June 2002 establishing a European Maritime Safety Agency [2002] OJ L 208/1, art 5(1): ‘The Agency shall be a body of the Community. It shall have legal personality.’; Regulation (EU) 2018/1971 of the European Parliament and of the Council of 11 December 2018 establishing the Body of European Regulators for Electronic Communications (BEREC) and the Agency for Support for BEREC (BEREC Office), amending Regulation (EU) 2015/2120 and repealing Regulation (EC) No 1211/2009 [2018] OJ L 321/1, art 2(1): ‘The BEREC Office shall be a body of the Union. It shall have legal personality.’ ↩︎
  92. See Council Regulation (EC) No 58/2003 of 19 December 2002 laying down the statute for executive agencies to be entrusted with certain tasks in the management of Community programmes [2003] OJ L 11/1, art 4(2): ‘An executive agency shall have legal personality. In each of the Member States, it shall enjoy the most extensive legal capacity accorded to legal persons under national law. It may, in particular, acquire or dispose of movable and immovable property and be a party to legal proceedings. To this end, it shall be represented by its Director.’; for an act establishing executive agencies and referencing Council Regulation (EC) No 58/2003, see Commission Implementing Decision (EU) 2021/173 of 12 February 2021 establishing the European Climate, Infrastructure and Environment Executive Agency, the European Health and Digital Executive Agency, the European Research Executive Agency, the European Innovation Council and SMEs Executive Agency, the European Research Council Executive Agency, and the European Education and Culture Executive Agency and repealing Implementing Decisions 2013/801/EU, 2013/771/EU, 2013/778/EU, 2013/779/EU, 2013/776/EU and 2013/770/EU [2021] OJ L 50/9, preamble. ↩︎
  93. See Parliament Amendments (n 12) Amendment 525, art 56(1): ‘The “European Artificial Intelligence Office” (the “ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ”) is hereby established. The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall be an independent body of the Union. It shall have legal personality.’ ↩︎
  94. The relevance of an express provision regarding an EU entity’s legal personality for determining the entity’s legal status has been recognised by the General Court in its case law on OLAF (see In vivo OOO v European Commission (n 90) para 14; Case T-658/17 Stichting Against Child Trafficking v European Commission [2018] ECLI:EU:T:2018:799, para 17). ↩︎
  95. See, with regard to OLAF, In vivo OOO v European Commission (n 90) paras 14–16. ↩︎
  96. See Wende (n 14) para 15 who points out that an enforcement measure taken by the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is to be regarded as a decision of the Commission, amenable to judicial review under article 263(4) TFEU; Bernsteiner and Schmitt (n 13) para 15 who state that the Commission is legally and politically responsible for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s actions; see also In vivo OOO v European Commission (n 90) paras 14–16 with regard to an action seeking to establish that OLAF failed to act (‘It should be noted that OLAF, established by Commission Decision 1999/352/EC, ECSC, Euratom of 28 April 1999 (OJ 1999 L 136, p. 20), is an internal department of the Commission, the independence of which is purely functional and limited to its investigation activities. Accordingly, and in the absence of any contrary provision, OLAF does not have separate legal personality and it is the Commission which acts on its behalf in legal proceedings. An action seeking, therefore, to establish that OLAF failed to act, in so far as it did not perform the external investigative functions laid down in Article 3 of Regulation (EC) No 1073/1999 of the European Parliament and of the Council of 25 May 1999 concerning investigations conducted by the European Anti-Fraud Office (OLAF) (OJ 1999 L 136, p. 1), as amended and applicable to the present proceedings, must be regarded as being directed against the Commission alone (see, to that effect and by analogy, judgment of 6 April 2006 in Camós Grau v Commission, T‑309/03, ECR, EU:T:2006:110, paragraph 66).’); see also with regard to OLAF Stichting Against Child Trafficking v European Commission (n 94) para 17; compare C-626/15 European Commission v Council of the European Union [2018] ECLI:EU:C:2018:925, paras 59–67 with regard to judicial review of measures adopted by Coreper, an auxiliary body of the Council, where the Council was the defendant. ↩︎
  97. See European Commission v Council of the European Union (n 96) para 59 (‘In accordance with settled case-law, any decision adopted by an institution, office, body or agency of the European Union, irrespective of its nature or form, which is intended to have legal effects constitutes a challengeable act for the purposes of Article 263 TFEU’). ↩︎
  98. See Section 3.1. and Section 3.2. respectively. ↩︎
  99. See Novelli and others (n 5) 575–577, 585–586 who discuss the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s institutional nature and conclude that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s ‘precise organisational structure […] and operational autonomy remain ambiguous’; Graux and others (n 62) 10 (‘[T]he AIO’s institutional position within the Commission raises concerns about its operational independence […].’). ↩︎
  100. See Establishment Decision (n 3) art 1(1). ↩︎
  101. See, for example, the DMA (n 15) and the DSA (n 15) which only refer to the Commission as a whole. ↩︎
  102. Parliament Amendments (n 12) Amendments 524–530; see Schmidl and Rohner (n 12) 983, s 3.1; Kirschke-Biller and Füllsack (n 12) para 546. ↩︎
  103. See Bernsteiner and Schmitt (n 13) para 14 who characterise the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as a non-independent part (‘unselbständige Einrichtung’) of the Commission; Wende (n 14) para 15 who emphasises that core decision competences lie with the Commission; see also Novelli and others (n 5) 575–577, 585–586 who, while expressing uncertainty about the extent of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s operational autonomy, state that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s autonomy is constrained by its incorporation into the Commission’s administrative structure; Roth-Isigkeit (n 6) para 1 who describes the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. as a particular institution (‘besondere Einrichtung’) without elaborating on whether this implies a special institutional status. Without express discussion of a special institutional status of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. : Hilgendorf and Härtlein (n 5) para 3; Schmidl and Rohner (n 12), 983 s 3.1. ↩︎
  104. Establishment Decision (n 3) art 1(2) and recitals 6 and 7. See Kirschke-Biller and Füllsack (n 12) para 553 who highlight the relevance of the Establishment Decision in the context of article 3(47)’s definition of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. . ↩︎
  105. Compare Regulation (EU, Euratom) No 883/2013 of the European Parliament and of the Council of 11 September 2013 concerning investigations conducted by the European Anti-Fraud Office (OLAF) and repealing Regulation (EC) No 1073/1999 of the European Parliament and of the Council and Council Regulation (Euratom) No 1074/1999 [2013] OJ L 248/1 (“OLAF Regulation”) art 3(1) (‘The Office [OLAF] shall exercise the power conferred on the Commission by Regulation (Euratom, EC) No 2185/96 to carry out on-the-spot checks and inspections in the Member States and, in accordance with the cooperation and mutual assistance agreements and any other legal instrument in force, in third countries and on the premises of international organisations.’ (emphasis added)) and art 17(3) (‘The Director-General [heading OLAF] shall neither seek nor take instructions from any government or any institution, body, office or agency in the performance of his duties with regard to the opening and carrying-out of external and internal investigations or to the drafting of reports following such investigations. If the Director-General considers that a measure taken by the Commission calls his independence into question, he shall immediately inform the Supervisory Committee, and shall decide whether to bring an action against the Commission before the Court of Justice.’). ↩︎
  106. Schmidl and Rohner (n 12) 983, s 3.1. ↩︎
  107. See in particular, AI Act, arts 3(47), 56(6), 88(1) and 90(2); see also Section 4.1. ↩︎
  108. For an analysis of article 88(1) and the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role in the supervision and enforcement of GPAI model provisions, see commentary on Article 88, Section 2.1.1. in this work. ↩︎
  109. See AI Act, arts 75(1) and 112(5) and recital 164. For an analysis of whether these provisions assign powers directly to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , see Section 4.1.3. ↩︎
  110. Schmidl and Rohner (n 12) 983, s 3.1. ↩︎
  111. Novelli and others (n 5) 575; Merkle (n 8) para 363. ↩︎
  112. Novelli and others (n 5) 575; see also Merkle (n 8) para 363 who solely mentions the European Parliament and the Council in this context. ↩︎
  113. Establishment Decision (n 3) art 6; see also Establishment Decision (n 3) art 2(3)(b) and (c). ↩︎
  114. See Establishment Decision (n 3) art 3(2)(h). The Establishment Decision (n 3) art 2(3)(c) does, however, expressly mention the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s cooperation with the European High Performance Computing Joint Undertaking (EuroHPC JU). ↩︎
  115. For the Commission’s interinstitutional cooperation, see, for example, Interinstitutional Agreement between the European Parliament, the Council of the European Union and the European Commission on Better Law-Making [2016] OJ L 123/1, paras 32–34. ↩︎
  116. See Novelli and others (n 5) 575 who recognise that ‘unlike interinstitutional services, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is integrated within the administrative framework of [DG CNECT]’. ↩︎
  117. See TFEU, art 265(1): ‘Should the European Parliament, the European Council, the Council, the Commission or the European Central Bank, in infringement of the Treaties, fail to act, the Member States and the other institutions of the Union may bring an action before the Court of Justice of the European Union to have the infringement established. This Article shall apply, under the same conditions, to bodies, offices and agencies of the Union which fail to act.’ (emphasis added). ↩︎
  118. See TFEU, art 267(1): ‘The Court of Justice of the European Union shall have jurisdiction to give preliminary rulings concerning: (a) the interpretation of the Treaties; (b) the validity and interpretation of acts of the institutions, bodies, offices or agencies of the Union’ (emphasis added). ↩︎
  119. See, for example, TEU, art 9 and TFEU, arts 15(1), 16(2) and 71. ↩︎
  120. See also TFEU, art 263(5): ‘Acts setting up bodies, offices and agencies of the Union may lay down specific conditions and arrangements concerning actions brought by natural or legal persons against acts of these bodies, offices or agencies intended to produce legal effects in relation to them.’ ↩︎
  121. See Section 3.1. ↩︎
  122. See TEU, art 13(1). ↩︎
  123. See, for example, TEU, art 9 (‘In all its activities, the Union shall observe the principle of the equality of its citizens, who shall receive equal attention from its institutions, bodies, offices and agencies.’) and TFEU, art 15(3) (‘Any citizen of the Union, and any natural or legal person residing or having its registered office in a Member State, shall have a right of access to documents of the Union’s institutions, bodies, offices and agencies […]’). One may note that the Treaties do not define the term ‘office of the Union’ (see Koen Lenaerts, Katheen Gutman and Janek Tomasz Nowak, EU Procedural Law (2nd edn, Oxford University Press, 2023) para 7.73 fn 263). ↩︎
  124. See Lenaerts, Gutman and Nowak (n 123) para 7.73 fn 263. One may note that other ‘offices’ – such as the European Asylum Support Office (EASO), the European Public Prosecutor’s Office (EPPO) and the Agency for Support for BEREC (BEREC Office) – have, notwithstanding their designation, been established not as offices but as Union bodies (see Regulation (EU) No 439/2010 of the European Parliament and of the Council of 19 May 2010 establishing a European Asylum Support Office [2010] OJ L 132/11, art 40(1): ‘The Support Office shall be a body of the Union. It shall have legal personality.’ (note that the European Asylum Support Office (EASO), established by Regulation (EU) No 439/2010 has been replaced and succeeded by the European Union Agency for Asylum under Regulation (EU) 2021/2303 of the European Parliament and of the Council of 15 December 2021 on the European Union Agency for Asylum and repealing Regulation (EU) No 439/2010 [2021] OJ L 468/1, art 1(1)); Council Regulation (EU) 2017/1939 (n 91) art 3(1): ‘The EPPO is hereby established as a body of the Union.’; Regulation (EU) 2018/1971 (n 91) art 2(1): ‘The BEREC Office shall be a body of the Union.’). ↩︎
  125. This provision stipulates that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall establish certain ‘forms of cooperation with bodies, offices and agencies of the Union’. The omission of ‘other’ before ‘bodies, offices and agencies of the Union’ contrasts with the wording of article 5(2) (‘other relevant Directorate-Generals’, emphasis added) and recital 7 (‘other Commission departments’) of the Establishment Decision (n 3) thereby reinforcing the conclusion that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is a Commission directorate rather than a body, office or agency of the Union. ↩︎
  126. Janine Wendt and Dominik Wendt, ‘§ 13 Aufsichtssystem’ in Janine Wendt and Dominik Wendt (eds), Das neue Recht der Künstlichen Intelligenz (Nomos 2025) para 10; opposing view: Wende (n 14) para 12. Article 2(27) of Regulation (EU, Euratom) 2024/2509 of the European Parliament and of the Council of 23 September 2024 on the financial rules applicable to the general budget of the Union (recast) [2024] OJ L 2024/2509 (“Financial Regulation”) defines a European office as ‘an administrative structure set up by the Commission, or by the Commission with one or more other Union institutions, to perform specific cross-cutting tasks’. ↩︎
  127. See Establishment Decision (n 3) recital 7, fn 6: ‘[The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ] is not a European Office within the meaning of the Financial Regulation, Regulation (EU, Euratom) 2018/1046 of the European Parliament and of the Council of 18 July 2018, Article 2(26).’ (As Regulation (EU, Euratom) 2018/1046 of the European Parliament and of the Council of 18 July 2018 on the financial rules applicable to the general budget of the Union, amending Regulations (EU) No 1296/2013, (EU) No 1301/2013, (EU) No 1303/2013, (EU) No 1304/2013, (EU) No 1309/2013, (EU) No 1316/2013, (EU) No 223/2014, (EU) No 283/2014, and Decision No 541/2014/EU and repealing Regulation (EU, Euratom) No 966/2012 [2018] OJ L 193/1 (“Old Financial Regulation”) has been recast by the Financial Regulation (n 126), the Establishment Decision’s reference to article 2(26) of the Old Financial Regulation must now be construed as a reference to the corresponding article 2(27) of the Financial Regulation (see Financial Regulation (n 126) arts 279 and 280 in conjunction with annex II). That the correlation table contained in annex II of the Financial Regulation does not account for the fact that the definition of a European office has moved from article 2(26) of the Old Financial Regulation to article 2(27) of the Financial Regulation appears to be a legislative oversight without any practical consequences.). ↩︎
  128. Schmidl and Rohner (n 12) 984, s 3.2; see Graux and others (n 62) 75; Novelli and others (n 5) 578. ↩︎
  129. Roth-Isigkeit (n 6) paras 9, 21 and 22; Paschke and Rachut (n 8) para 5; Schmidl and Rohner (n 12) 983, s 3.1; see also Merkle (n 8) para 364 emphasising the role of the Establishment Decision (n 3). ↩︎
  130. See also AI Act, recital 148; Establishment Decision (n 3) recitals 4 and 5; see further Schmidl and Rohner (n 12) 983, s 3.1. ↩︎
  131. Establishment Decision (n 3) arts 2(1) and 3; see Section 4.1. ↩︎
  132. Establishment Decision (n 3) art 2(2); see Graux and others (n 62) 75; Novelli and others (n 5) 578; Section 4.2. ↩︎
  133. See Section 4.1. and Section 4.2. ↩︎
  134. See Novelli and others (n 5) 578. ↩︎
  135. See Novelli and others (n 5) 578 who base their view on article 2(1) of the Establishment Decision (n 3). However, this does not refer to the harmonised implementation and enforcement of the AI Act but rather to the implementation and enforcement of ‘the forthcoming Regulation laying down harmonised rules on artificial intelligence’, that is, the AI Act. ↩︎
  136. Paschke and Rachut (n 8) para 11; similar: Wende (n 14) para 1 who emphasises the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s coordinative tasks. ↩︎
  137. Roth-Isigkeit (n 6) para 3. ↩︎
  138. See Establishment Decision (n 3) arts 2(1) and (2), and 3(1). ↩︎
  139. See also AI Act, art 3(47): ‘“ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ” means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. and general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. , and AI governance’. ↩︎
  140. See Establishment Decision (n 3) arts 2(1) and 3; Schmidl and Rohner (n 12) 984, s 3.2. ↩︎
  141. See Establishment Decision (n 3) arts 2(1), 3(1) and (2) which expressly refer to ‘the forthcoming Regulation’, that is, the AI Act, with regard to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s tasks. For the AI Act’s governance framework, see, in particular, AI Act, recitals 148 and 164. ↩︎
  142. See AI Act, art 74(1). ↩︎
  143. See, in particular, AI Act, ch V. ↩︎
  144. See AI Act, arts 88(1) and 101; for the Commission’s power to designate GPAI models as presenting systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. and to decide about providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. ’ challenges to classification of their models as presenting systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. , see AI Act, arts 51(1)(b) and 52. ↩︎
  145. See AI Act, art 75(1); see also forthcoming commentary on Article 75(1) in this work. ↩︎
  146. For delegated acts with particular relevance to GPAI model governance, see AI Act, arts 51(3), 52(4), 53(5) and (6) (on these provisions, see commentary on Article 51, Section 2.3. in this work; commentary on Article 52, Section 2.3.2. in this work; and commentary on Article 53, Sections 2.5. and 2.6. in this work respectively); for further delegated acts, see AI Act, arts 6(6) and (7), 7(1) and (3), 11(3), 43(5) and (6), and 47(5). For the Commission’s exercise of the AI Act’s delegations, see AI Act, art 97. ↩︎
  147. For implementing acts with particular relevance to GPAI model governance, see AI Act, art 56(6) and (9) (see commentary on Article 56, Section 2.9.2. in this work), 68(1) and (5) (see commentary on Article 68, Section 2.1.1. in this work), 92(6) and 101(6); for further implementing acts, see AI Act, arts 37(4), 41, 50(7), 58, 60(1), 69(2) and 72(3). ↩︎
  148. See AI Act, art 96. For guidelines with particular relevance to GPAI model governance, see Commission Guidelines (n 66). ↩︎
  149. For codes of practice with particular relevance to GPAI model governance, see AI Act, art 56 (see the commentary on Article 56 in this work); for further codes of practice, see AI Act, art 50(7). ↩︎
  150. See AI Act, art 95. ↩︎
  151. Establishment Decision (n 3) art 2(1). ↩︎
  152. See Roth-Isigkeit (n 6) paras 21 and 22 who distinguishes between tasks under the AI Act (‘Aufgaben aus der Verordnung’) and tasks upon assignment (‘Aufgaben nach Beauftragung’); see also Schmidl and Rohner (n 12) 986, s 3.2 who argue that the tasks under article 3(1)(a)–(d) of the Establishment Decision (n 3) ‘are ones actually stemming directly from the AI Act’. ↩︎
  153. The French and German language versions of article 3(2) of the Establishment Decision (n 3) do not support the conclusion that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. may not carry out the tasks contained in article 3(2) of the Establishment Decision (n 3) on its own initiative, as the English language version might suggest. Their use of ‘est chargé’ and ‘wird […] beauftragt’ does not imply the need for a specific assignment or request (see Establishment Decision (n 3) art 3(2) (German language version): ‘Im Hinblick auf die wirksame Durchführung der künftigen Verordnung wird das Amt darüber hinaus beauftragt, [..]’; Establishment Decision (n 3), art 3(2) (French language version): ‘En outre, afin de contribuer à la mise en œuvre effective du futur règlement, le Bureau est chargé: […]’. Moreover, article 3(2) of the Establishment Decision (n 3) does not expressly state who should assign its tasks to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , and one task it contains – ‘encouraging and facilitating the drawing up of codes of practices’ (Establishment Decision (n 3) art 3(2)(i)) – is a task that the AI Act, by its wording, directly confers upon the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. (see AI Act, art 56(1): ‘The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. shall encourage and facilitate the drawing up of codes of practice at Union level in order to contribute to the proper application of this Regulation, taking into account international approaches.’), which casts further doubt on article 3(2) of the Establishment Decision tasks always requiring a prior assignment. ↩︎
  154. Schmidl and Rohner (n 12) 986, s 3.2 suggest that the tasks in article 3(2) of the Establishment Decision are of an ‘assisting, coordinating and harmonizing nature’. This appropriately characterizes the tasks contained in article 3(2) of the Establishment Decision. It does not, however, explain well the distinction between article 3(1) and (2) of the Establishment Decision as points (e) and (f) of article 3(1) of the Establishment Decision contain assisting and coordinating tasks as well. Moreover, Merkle (n 8) paras 366, 368 suggests that the tasks contained in article 3(1) of the Establishment Decision primarily relate to GPAI models and GPAI systems, whereas article 3(2) of the Establishment Decision contains further tasks (similar: Wendt and Wendt (n 126) paras 12, 16). However, while it is true that many of the tasks under article 3(1) of the Establishment Decision concern GPAI models and GPAI systems, point (f) of article 3(1) of the Establishment Decision demonstrates that such a connection to GPAI models and GPAI systems is not necessarily required. Moreover, many of the tasks under article 3(2) of the Establishment Decision appear to be of relevance with regard to GPAI model and GPAI system governance as well. ↩︎
  155. Schmidl and Rohner (n 12) 986, s 3.2.1; see Roth-Isigkeit (n 6) paras 22 and 23 who emphasises that these provisions specify the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s tasks concerning GPAI models under the AI Act. The other two points, article 3(1)(e) and (f) of the Establishment Decision (n 3), concern the rules on AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. and prohibited AI practices (for an analysis of the relevant tasks, see Roth-Isigkeit (n 6) paras 30–34; Schmidl and Rohner (n 12) 986, s 3.2.1). ↩︎
  156. Establishment Decision (n 3) art 3(1)(a). ↩︎
  157. For these assessment instruments, see commentary on Article 51, Section 2.1.1.3. in this work. See, however, Roth-Isigkeit (n 6) para 27 who primarily relates this task to the substantive obligations for providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of GPAI models with systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. under article 55. ↩︎
  158. See commentary on Article 51, Section 2.1.2.1.2. in this work. ↩︎
  159. For article 51(3)’s delegation of power, see commentary on Article 51, Section 2.3. in this work; for article 52(4)’s delegation of power, see commentary on Article 52, Section 2.3.2. in this work. ↩︎
  160. See AI Act, art 91(1): ‘The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , after consulting the Board, may conduct evaluations of the general-purpose AI model Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. concerned: (a) to assess compliance of the provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. with obligations under this Regulation, where the information gathered pursuant to Article 91 is insufficient; or, (b) to investigate systemic risks Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. at Union level of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. with systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. , in particular following a qualified alert from the scientific panel in accordance with Article 90(1), point (a).’ The Commission may also ‘appoint independent experts to carry out evaluations on its behalf’ pursuant to article 91(2). ↩︎
  161. See commentary on Article 51, Section 2.2. in this work. ↩︎
  162. Establishment Decision (n 3) art 3(1)(b). ↩︎
  163. For an analysis of article 75(1), see forthcoming commentary on Article 75(1) in this work. ↩︎
  164. See Roth-Isigkeit (n 6) paras 24–27 who connects this task to the substantive GPAI model provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. obligations under articles 53–55. ↩︎
  165. See also AI Act, recital 162, second sentence: ‘The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. should be able to carry out all necessary actions to monitor the effective implementation of this Regulation as regards general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. .’ ↩︎
  166. Establishment Decision (n 3) art 3(1)(c). For the corresponding AI Act provisions on the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s receipt of qualified alerts from the scientific panel, see AI Act, art 90(1) and (2). ↩︎
  167. AI Act, art 90(1). ↩︎
  168. AI Act, art 90(2). ↩︎
  169. See commentary on Article 52, Section 2.3.1. in this work. ↩︎
  170. Establishment Decision (n 3) art 3(1)(d). ↩︎
  171. For the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role in the oversight of GPAI models, see commentary on Article 88, Section 2.1.3. in this work. ↩︎
  172. See commentary on Article 88, Section 2.1.4. in this work. ↩︎
  173. For an analysis of the Commission’s power to request measures under article 93, see commentary on Article 93 in this work. ↩︎
  174. See AI Act, arts 3(68), 25(1)(c), 50(2) and 75(2). ↩︎
  175. See AI Act, art 74(1). ↩︎
  176. For the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s role in supporting the Commission, see Schmidl and Rohner (n 12) 986, s 3.2.2; Roth-Isigkeit (n 6) para 36. ↩︎
  177. Establishment Decision (n 3) art 3(2)(a). For Commission decisions in the context of supervision and enforcement of GPAI model obligations, see AI Act, arts 91(1), 92(3) and 93(1) (see commentary on Article 93 in this work). For Commission decisions in the context of systemic risk Article 3(65) AI Act: ‘systemic risk’ means a risk that is specific to the high-impact capabilities of general-purpose AI models, having a significant impact on the Union market due to their reach, or due to actual or reasonably foreseeable negative effects on public health, safety, public security, fundamental rights, or the society as a whole, that can be propagated at scale across the value chain. classification of GPAI models, see AI Act, art 51(1)(b) (see commentary on Article 51, Section 2.1.2.) and AI Act, art 52 (see commentary on Article 52 in this work). For delegated acts with particular relevance to GPAI model governance, see AI Act, arts 51(3), 52(4), 53(5) and (6) (see commentary on Article 51, Section 2.3. in this work; commentary on Article 52, Section 2.3.2. in this work and commentary on Article 53, Sections 2.5. and 2.6. in this work respectively). For implementing acts with particular relevance to GPAI model governance, see AI Act, art 56(6) and (9) (see commentary on Article 56, Section 2.9.1. in this work), 68(1) and (5) (see commentary on Article 68 in this work), 92(6) and 101(6). ↩︎
  178. Establishment Decision (n 3) art 3(2)(c). For guidelines of particular relevance to GPAI model governance, see Commission Guidelines (n 66). ↩︎
  179. Establishment Decision (n 3) art 3(2)(d); see Herbers and Rappenglück (n 6) para 24. For standardisation requests, see AI Act, art 40. ↩︎
  180. See AI Act, art 96(1): ‘The Commission shall develop guidelines on the practical implementation of this Regulation, and in particular on: (a) the application of the requirements and obligations referred to in Articles 8 to 15 and in Article 25; (b) the prohibited practices referred to in Article 5; (c) the practical implementation of the provisions related to substantial modification Article 3(23) AI Act: ‘substantial modification’ means a change to an AI system after its placing on the market or putting into service which is not foreseen or planned in the initial conformity assessment carried out by the provider and as a result of which the compliance of the AI system with the requirements set out in Chapter III, Section 2 is affected or results in a modification to the intended purpose for which the AI system has been assessed. ; (d) the practical implementation of transparency obligations laid down in Article 50; (e) detailed information on the relationship of this Regulation with the Union harmonisation legislation listed in Annex I, as well as with other relevant Union law, including as regards consistency in their enforcement; (f) the application of the definition of an AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. as set out in Article 3, point (1).’ ↩︎
  181. Clara Saillant and Katarzyna Barud, ‘Article 96 Guidelines from the Commission on the Implementation of this Regulation’ in Ceyhun Necati Pehlivan, Nikolaus Forgó and Peggy Valcke (eds), The EU Artificial Intelligence (AI) Act: A Commentary (Wolters Kluwer 2024) 1327, s 1; Sarah Hartmann, ‘Art. 96 Leitlinien der Kommission zur Durchführung dieser Verordnung’ in Mario Martini and Christiane Wendehorst (eds), KI-VO: Verordnung über Künstliche Intelligenz: Kommentar (2nd edn, C H Beck 2026) para 6; Christina Brandt-Steinke, ‘Art. 96: Leitlinien der Kommission zur Durchführung dieser Verordnung’ in Jens Schefzig and Robert Kilian (eds), Beck’scher Online-Kommentar KI-Recht (4th edn, C H Beck 2025) para 2. ↩︎
  182. For the Commission’s general approach of ensuring the application of Union law, inter alia, through the provision of guidance, see European Commission, ‘Communication from the Commission: Enforcing EU law for a Europe that delivers’ COM (2022) 518 final, 10–11: ‘To avoid diverging interpretations of newly adopted EU law and to foster a common understanding of existing rules, the Commission provides practical guidance to Member States, businesses, stakeholders, and the public on how to understand and apply specific aspects of EU law. This guidance can, for instance, take the form of guidelines on the interpretation and application of EU law or “Frequently Asked Questions” published online. They have been issued in all major policy areas. In particular, the Commission uses guidelines to accompany Member States through the transposition process for directives and application of regulations, usually starting right after the adoption of an act, as well as to consolidate the case-law of the Court of Justice.’; for Commission guidance in the field of EU competition law, Zlatina Georgieva, ‘Soft Law in EU Competition Law and its Judicial Reception in Member States: A Theoretical Perspective’ (2019) 16 German Law Journal 223, 223; for a typology of Commission guidance documents distinguishing between (i) interpretative guidance, (ii) implementing guidance, (iii) explanatory guidance, (iv) technical guidance and (v) the dissemination of good practices, see Clara van Dam, ‘Guidance Documents of the European Commission: A Typology to Trace the Effects in the National Legal Order’ (2017) 10 Review of European Administrative Law 75, 82–90. ↩︎
  183. See, for example, Communication from the Commission — Notice on agreements of minor importance which do not appreciably restrict competition under Article 101(1) of the Treaty on the Functioning of the European Union (De Minimis Notice) [2014] OJ C 291/1. ↩︎
  184. See, for an overview of templates in the field of state aid law, European Commission, ‘Forms for Notifications and Reporting’ <https://competition-policy.ec.europa.eu/state-aid/legislation/forms-notifications-and-reporting_en> accessed 25 January 2026. ↩︎
  185. See Commission Notice on informal guidance relating to novel or unresolved questions concerning Articles 101 and 102 of the Treaty on the Functioning of the European Union that arise in individual cases (guidance letters) [2022] OJ C 381/9. ↩︎
  186. See, for example, European Commission, ‘Comfort letter: cooperation at a Matchmaking Event – Towards COVID19 vaccines upscale production’ (2021) COMP/E-1/GV/BV/nb (2021/034137). ↩︎
  187. See, for example, European Commission, ‘Questions and Answers: Digital Markets Act: Ensuring fair and open digital Markets’ (2023) https://ec.europa.eu/commission/presscorner/api/files/document/print/en/qanda_20_2349/QANDA_20_2349_EN.pdf accessed 25 January 2026. ↩︎
  188. Georgieva (n 182) 226 (‘Article 17.1 TEU authorizes the Commission to produce soft law […].’); see also European Commission, ‘Communication from the Commission: Enforcing EU law for a Europe that delivers’ (n 182) 1, 10–11. ↩︎
  189. Establishment Decision (n 3) art 3(2)(b); see Roth-Isigkeit (n 6) para 37; Schmidl and Rohner (n 12) 988, s 3.2.6 (describing this task as an ‘overarching task’). ↩︎
  190. Establishment Decision (n 3) art 3(2)(g). This includes ‘preparing the set-up of advisory bodies at Union level’ (Establishment Decision (n 3) art 3(2)(g)). ↩︎
  191. Establishment Decision (n 3) art 3(2)(h); see AI Act, art 65(8), second sentence. ↩︎
  192. Establishment Decision (n 3) art 3(2)(h). This includes ‘providing the administrative set-up, organising meetings and preparing relevant documents’ (Establishment Decision (n 3) art 3(2)(h)); see AI Act, arts 67 and 68. See also Schmidl and Rohner (n 12) 987, s 3.2.3. ↩︎
  193. Establishment Decision (n 3) art 3(2)(i). For the corresponding AI Act provisions, see AI Act, arts 50 and 56 (concerning codes of practice) and AI Act, art 95 (concerning codes of conduct). ↩︎
  194. Establishment Decision (n 3) art 3(2)(e); see Schmidl and Rohner (n 12) 988, s 3.2.5. For the corresponding AI Act provisions on regulatory sandboxes, see AI Act, arts 57–61. The AI Act only foresees regulatory sandboxes on the Member State, regional and local level (see AI Act, art 57(1) and (2)) which therefore do not encompass GPAI model provisions overseen at the Union level (see AI Act, art 88(1)). ↩︎
  195. Establishment Decision (n 3) art 3(2)(f). ↩︎
  196. For an overview over the Commission, both in its meaning as the College of Commissioners and the permanent bureaucracy, see Paul Craig and Gráinne de Búrca, EU Law: Text, Cases, and Materials (7th edn, Oxford University Press 2020) 61–70; Christian Calliess, ‘The Internal Organisation of the European Commission: Between Efficiency and Politicisation’ (Berliner Online-Beiträge zum Europarecht, Nr. 152, 2024) <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5105881> accessed 9 January 2025, 3–15. ↩︎
  197. See, in particular, AI Act, arts 51(3), 52(4), 53(5) and (6). For the conditions regarding the Commission’s exercise of its power to adopt delegated acts, see AI Act, art 97. ↩︎
  198. See, in particular, AI Act, arts 41, 56(6) and (9), 68(1), 92(6), and 101(6). ↩︎
  199. See AI Act, arts 51(1)(b) and 52(1), third sentence and 52(4), first subparagraph. ↩︎
  200. See AI Act, arts 91(1), 92(3) and 93(1). ↩︎
  201. See AI Act, art 101(1). ↩︎
  202. See Wende (n 14) para 15 who points out that core decision competences concerning the AI Act’s implementation lie with the Commission; Bernsteiner and Schmitt (n 13) para 15 who lay out that the fact that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has been afforded tasks is without prejudice to the principle of collegiate responsibility governing the Commission’s actions and that the Commission remains politically and legally responsible for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s actions. Quentin B. Schäfer, ‘Article 88: Enforcement of the Obligations of Providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of General-Purpose AI Models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. ’ in Ceyhun Necati Pehlivan, Nikolaus Forgó and Peggy Valcke (eds), The EU Artificial Intelligence (AI) Act: A Commentary (Wolters Kluwer 2024) 1236, s 1 (with regard to the enforcement of GPAI model obligations); see also AI Act, art 91(1) and (2) (envisaging that the Commission takes the decision to request information from GPAI model providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. while the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ‘send[s]’ the request for information); see, however, Schmidl and Rohner (n 12) 984, s 3.2 who argue that ‘since the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is part of the Commission, all tasks of the Commission specified in the AI Act may be handled by the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’ without discussing whether this also applies to the issuance of legally binding acts. ↩︎
  203. See Commission Rules of Procedure (n 11) art 45(2): ‘The mission of all services [including Directorates-General] shall be to assist the Commission in the preparation and performance of its tasks, and in the achievement of its priorities.’; for the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s nature as a Commission directorate, see Section 3.1. ↩︎
  204. See Section 3.2. ↩︎
  205. For the principle of collegiate responsibility, see fn 216. ↩︎
  206. See, in particular, AI Act, arts 3(47), 56(6), 88(1) and 90(2). ↩︎
  207. See, in particular, Establishment Decision (n 3) art 3(2)(a), (c) and (d). ↩︎
  208. See Commission Guidelines (n 66) para 90: ‘[T]he Commission, acting through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , is entrusted with supervising and enforcing [GPAI model] obligations’; see also Commission Guidelines (n 66) para 103 concerning the Commission’s enforcement powers. ↩︎
  209. See commentary on Article 88, Section 2.1.4.; Schäfer (n 202) 1236, s 1. Similarly, article 90(2) provides that, upon a qualified alert of the scientific panel, the Commission may exercise its powers ‘through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’. And under article 56(6), it is the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. that assesses the adequacy of codes of practice and the Commission that approves them by way of an implementing act. More generally, where the AI Act provides for the issuance of legally binding acts, such as delegated acts (see, in particular, AI Act, art 97(1)), implementing acts (AI Act, arts 92(6) and 101(6)), designation decisions (AI Act, arts 51(1)(b), 52(1) and (4)), requests vis-à-vis GPAI model providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. (AI Act, arts 91(1), 92(3) and 93(1)), or the imposition of fines (AI Act, art 101(1)), it commonly refers to the Commission. Where preparatory, implementing and supporting tasks related to those legally binding acts are in question, the AI Act commonly refers to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. . For example, this is the case for monitoring GPAI model provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. compliance (AI Act, art 89(1)), sending requests for information issued by the Commission (AI Act, art 91(2)), conducting GPAI model evaluations (AI Act, art 92(1)), or encouraging and facilitating the drawing up of codes of practice (AI Act, art 56(1)). ↩︎
  210. See article 112(5) which – in the context of evaluation of the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s functioning – refers to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s ‘powers and competences’ and ‘enforcement competences’. Moreover, recital 164 states that ‘[t]he AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. should be able to investigate possible infringements in accordance with the powers provided for in this Regulation, including by requesting documentation and information, by conducting evaluations, as well as by requesting measures from providers Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. .’ ↩︎
  211. To the same effect, specifically mentioning article 75(1): Wende (n 14) para 15. ↩︎
  212. See also AI Act, recital 161, second sentence: ‘To avoid overlapping competences, where an AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. is based on a general-purpose AI model Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and the model and system are provided by the same provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. , the supervision should take place at Union level through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , which should have the powers of a market surveillance authority Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. within the meaning of Regulation (EU) 2019/1020 for this purpose.’ ↩︎
  213. See Commission Rules of Procedure (n 11) art 45(2), according to which ‘[t]he mission of all services [including Directorates-General] shall be to assist the Commission in the preparation and performance of its tasks, and in the achievement of its priorities.’ (emphasis added). ↩︎
  214. OLAF, which is headed by a Director-General (OLAF Regulation (n 105) art 17(1)) and has been characterized by the General Court as an ‘internal department of the Commission’ (In vivo OOO v European Commission (n 90) para 14), enjoys independence from the Commission in the exercise of its investigative function (see OLAF Regulation (n 105) art 17(3) and recital 3; see also In vivo OOO v European Commission (n 90) para 14). ↩︎
  215. See Schmidl and Rohner (n 12) 983, s 3.1 who does not accord any interpretive significance to the AI Act’s use of the terms AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. and Commission in the AI Act, contending that the AI Act’s use of both terms is a ‘result of the political negotiations and the late changes to the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’. ↩︎
  216. The principle of collegiate responsibility is ‘founded on the equal participation of the members of the Commission in the adoption of decisions and it follows from that principle, in particular, that decisions should be subject Article 3(58) AI Act: ‘subject’, for the purpose of real-world testing, means a natural person who participates in testing in real-world conditions. of a collective deliberation and that all the members of the College of Commissioners bear collective responsibility on the political level for all decisions adopted’ (Case C-5/85 AKZO Chemie BV and AKZO Chemie UK Ltd v Commission of the European Communities [1986] ECLI:EU:C:1986:328, para 30; for an overview over the European courts’ jurisprudence on the principle of collegiate responsibility, see Maria Patrin, The Principle of Collegiality in the Commission’s Decision-Making: Legal Substance and Institutional Practice (PhD Thesis, European University Institute, 2020) 100–116). ↩︎
  217. See AI Act, recital 161, second sentence: ‘To avoid overlapping competences, where an AI system Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. is based on a general-purpose AI model Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and the model and system are provided by the same provider Article 3(3) AI Act: ‘provider’ means a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge. , the supervision should take place at Union level through the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. , which should have the powers of a market surveillance authority Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. within the meaning of Regulation (EU) 2019/1020 for this purpose.’ ↩︎
  218. See OLAF Regulation (n 105) art 1(1): ‘In order to step up the fight against fraud, corruption and any other illegal activity affecting the financial interests of the European Union and of the European Atomic Energy Community (hereinafter referred to collectively, when the context so requires, as “the Union”), the European Anti-Fraud Office established by Decision 1999/352/EC, ECSC, Euratom (“the Office”) shall exercise the powers of investigation conferred on the Commission by: (a) the relevant Union acts; and (b) the relevant cooperation and mutual assistance agreements concluded by the Union with third countries and international organisations.’ (emphasis added); see also OLAF Regulation (n 105) recital 3: ‘The Commission has given the Office full independence to exercise its investigative function.’ ↩︎
  219. See Establishment Decision (n 3) art 3(2)(a) setting out that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. is (merely) tasked with ‘assisting the Commission in the preparation of relevant Commission Decisions, and of implementing and delegated acts’ (emphasis added). ↩︎
  220. See commentary on Article 88, Section 2.1.4. in this work. ↩︎
  221. See Roth-Isigkeit (n 6) para 39; Merkle (n 8) para 364 who calls them general tasks (‘allgemeine Aufgaben’); Wendt and Wendt (n 126) para 16. ↩︎
  222. For an overview, see Roth-Isigkeit (n 6) paras 39–43. ↩︎
  223. See Establishment Decision (n 3) art 2(2): ‘The Office shall have the following additional tasks: (a) contribute to the strategic, coherent and effective Union approach to international initiatives on AI pursuant to Article 7 in coordination with Member States and in line with Union positions and policies’; see also Establishment Decision (n 3) art 7: ‘The Office shall contribute to international cooperation on AI by the Commission, including innovation and excellence policy, with third countries and international organisations, in particular by: (a) advocating the responsible stewardship of AI and promoting the Union approach to trustworthy AI; (b) contributing to international cooperation related to AI regulation and governance; (c) contributing to the implementation of international agreements on rules on AI, including by providing support to Member States.’ ↩︎
  224. See Establishment Decision (n 3) art 2(2)(b): ‘[The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has the additional task to] contribute to fostering actions and policies in the Commission that reap the societal and economic benefits of AI technologies pursuant to Article 5 [of the Establishment Decision].’ The Establishment Decision envisages the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. ’s fulfillment of this task primarily through its cooperation within the Commission, as evidenced by article 5 of the Establishment Decision (n 3) entitled ‘Cross-sectoral cooperation within the Commission’ (see Establishment Decision (n 3) art 5: ‘1. The Office shall contribute to fostering actions and policies that reap the societal and industrial benefit of AI technologies. 2. In particular, the Office shall (a) work with other relevant Directorate-Generals and services of the Commission in the performance of its tasks pursuant to Article 2, notably with the European Centre for Algorithmic Transparency as regards the evaluation and testing of general-purpose AI models Article 3(63) AI Act: ‘general-purpose AI model’ means an AI model, including where such an AI model is trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market and that can be integrated into a variety of downstream systems or applications, except AI models that are used for research, development or prototyping activities before they are placed on the market. and systems; (b) support other relevant Directorate-Generals and services of the Commission with a view to facilitate the use of AI models and systems as transformative tools in the relevant domains of Union policies, as well as to raise awareness about emerging risks Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. .’). ↩︎
  225. See Establishment Decision (n 3) art 2(2)(c): ‘[The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has the additional task to] support the accelerated development, roll-out and use of trustworthy AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. and applications that bring societal and economic benefits and that contribute to the competitiveness and the economic growth of the Union. In particular, the Office shall promote innovation ecosystems by working with relevant public and private actors and the startup community’. ↩︎
  226. Establishment Decision (n 3) art 2(2)(d): ‘[The AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. has the additional task to] monitor the evolution of AI markets and technologies.’ ↩︎
  227. Establishment Decision (n 3) art 7, point (a). ↩︎
  228. Establishment Decision (n 3) art 7, points (b) and (c). Recital 4 of the Establishment Decision (n 3) mentions the G7 Code of Conduct and Guiding Principles for developers of advanced AI systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. as examples of international AI rules and principles. ↩︎
  229. European Commission, ‘Fostering global AI safety: EU AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. participates in inaugural International Network of AI Safety Institutes meeting’ (2024) <https://digital-strategy.ec.europa.eu/en/news/fostering-global-ai-safety-eu-ai-office-participates-inaugural-international-network-ai-safety> accessed 9 January 2025. ↩︎
  230. International Network of AI Safety Institutes, ‘Mission Statement’ (2024) <https://www.nist.gov/system/files/documents/2024/11/20/Mission%20Statement%20-%20International%20Network%20of%20AISIs.pdf> accessed 9 January 2025; see also International Network of AI Safety Institutes ‘Joint Statement on Risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. Assessment of Advanced AI Systems Article 3(1) AI Act: ‘AI system’ means a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. ’ (2024) (accessible under <https://digital-strategy.ec.europa.eu/en/news/fostering-global-ai-safety-eu-ai-office-participates-inaugural-international-network-ai-safety> accessed 9 January 2025). ↩︎
  231. European Commission, ‘ AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. contributes to the third-joint testing exercise of the International Network of AI Safety Institutes’ (2025) <https://digital-strategy.ec.europa.eu/en/news/ai-office-contributes-third-joint-testing-exercise-international-network-ai-safety-institutes> accessed 9 January 2025. ↩︎
  232. See Paschke and Rachut (n 8) para 10. ↩︎
  233. See Section 3.1. ↩︎
  234. Roth-Isigkeit (n 6) para 14; Novelli and others (n 5) 575. For the Commission’s rules of procedure, see Commission Rules of Procedure (n 11). ↩︎
  235. See Roth-Isigkeit (n 6) para 14; Schmidl and Rohner (n 12) 987–988, s 3.2.4; Merkle (n 8) para 369. ↩︎
  236. Establishment Decision (n 3) arts 2(3)(a) and 4; see Roth-Isigkeit (n 6) para 18; Schmidl and Rohner (n 12) 987, s 3.2.4. ↩︎
  237. Establishment Decision (n 3) arts 2(3)(b) and 5; see Roth-Isigkeit (n 6) paras 15 and 16 who argues on the basis of article 5(2)(b) of the Establishment Decision (n 3) that the main focus of this cooperation within the Commission is the Commission’s forward-looking activities that extend beyond AI’s product safety law dimension; Schmidl and Rohner (n 12) 987–988, s 3.2.4. ↩︎
  238. Establishment Decision (n 3) arts 2(3)(c) and 6; see Novelli and others (n 5) 575 highlighting that the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. resembles in this respect interinstitutional services such as the EU’s Computer Emergency Response Team (CERT-EU); Roth-Isigkeit (n 6) para 19 who argues that the Establishment Decision aims to ensure that this cooperation does not take place merely on an ad hoc basis; Schmidl and Rohner (n 12) 988, s 3.2.4. ↩︎
  239. Establishment Decision (n 3) arts 2(3)(d) and 7(1)(c); for the cooperation between the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. and Member States, see also AI Act, art 64(2). ↩︎
  240. Establishment Decision (n 3) art 7; see Roth-Isigkeit (n 6) para 20; Schmidl and Rohner (n 12) 988, s 3.2.4. ↩︎
  241. Establishment Decision (n 3) recital 7; see Paschke and Rachut 64 (n 8) para 10. ↩︎
  242. See AI Act, art 78(1): ‘1. The Commission, market surveillance authorities Article 3(26) AI Act: ‘market surveillance authority’ means the national authority carrying out the activities and taking the measures pursuant to Regulation (EU) 2019/1020. and notified bodies Article 3(22) AI Act: ‘notified body’ means a conformity assessment body notified in accordance with this Regulation and other relevant Union harmonisation legislation. and any other natural or legal person involved in the application of this Regulation shall, in accordance with Union or national law, respect the confidentiality of information and data obtained in carrying out their tasks and activities in such a manner as to protect, in particular:
    (a) the intellectual property rights and confidential business information or trade secrets of a natural or legal person, including source code, except in the cases referred to in Article 5 of Directive (EU) 2016/943 of the European Parliament and of the Council (57);
    (b) the effective implementation of this Regulation, in particular for the purposes of inspections, investigations or audits;
    (c) public and national security interests;
    (d) the conduct of criminal or administrative proceedings;
    (e) information classified pursuant to Union or national law.’ ↩︎
  243. AI Act, recital 161: ‘In order to ensure trustful and constructive cooperation of competent authorities on Union and national level, all parties involved in the application of this Regulation should respect the confidentiality of information and data obtained in carrying out their tasks, in accordance with Union or national law’. ↩︎
  244. AI Act, art 112(5); see Schmidl and Rohner (n 12) 989, s 3.3. ↩︎
  245. AI Act, art 112(5). ↩︎
  246. AI Act, art 112(9). ↩︎
  247. AI Act, art 112(8). ↩︎
  248. AI Act, art 112(10). ↩︎
  249. AI Act, art 112(13). ↩︎
  250. AI Act, art 112(10). ↩︎