Cambridge Commentary on EU General-Purpose AI Law

Commentary

1. General remarks

1The scientific panel of independent experts forms part of the broader governance framework that aims to support the application of the AI Act, build capabilities at an EU level, and integrate stakeholders, including the scientific community.¹ The purpose of the panel is to support the enforcement activities under the AI Act,² in particular focusing on the monitoring activities 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 regards general-purpose AI (“GPAI”) models.³ As scholars have pointed out, information asymmetries pose a central challenge in AI governance: ‘Private industry that develops AI may learn about emergent risks Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. , but government currently lacks the ability to identify, verify, and act on such risks Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. as they emerge’.⁴ In this light, the scientific panel can be seen as an evidence-seeking body that supports 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 identifying, verifying and acting on (systemic) 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 harms from GPAI models as they emerge.⁵ From a broader risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. governance perspective, knowledge is essential for the legitimacy of the regulator as an authority, including as measured by the credibility of the expert bodies they consult.⁶ Thus, the scientific panel also serves to strengthen the legitimacy 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. and national 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. (“MSAs”).

2Article 68 is located within Section 1. of Chapter VII AI Act, titled ‘Governance at Union level’. Key actions of the scientific panel are laid down in separate provisions in Section 5 of Chapter VII, titled ‘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. ’. These include providing qualified alerts 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. under Article 90 AI Act⁷ and requesting the Commission to require documentation or information from a 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. in accordance with Article 91(3) AI Act.⁸ Further, independent experts from the panel may be appointed by the Commission to carry out evaluations on its behalf in accordance with Article 92(2) AI Act.⁹ That being said, the scientific panel does not have powers to 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. or the Commission to take specific actions, it merely ‘advises and supports’.¹⁰

3The Implementing Regulation¹¹ establishing the scientific panel (“Implementing Act”) entered into force on 30 March 2025.¹² As part of the drafting procedure, there was an open feedback process,¹³ and the provisions of the Implementing Act were scrutinised and approved by the Artificial Intelligence Committee, consisting of Member States representatives, in accordance with the comitology procedure as required by Article 68(1) AI Act.¹⁴ The Implementing Act includes details on the panel, its constitution, and its operation. In particular, it elaborates on the ‘conditions, procedures and detailed arrangements’ related to raising qualified alerts and requesting assistance 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. for performing tasks, as required by Article 68(5) AI Act.¹⁵ However, many details remain unaddressed in the Implementing Act. These will likely be fleshed out once the scientific panel is constituted and adopts rules of procedure, which is yet to occur as of the time of writing (September 2025).

4The structure of Article 68 is somewhat disjointed. The first paragraph addresses the function of the scientific panel and its establishment by means of an implementing act. The second paragraph concerns the composition of the panel. The third paragraph details the tasks of the panel. The fourth paragraph concerns the independence of panel experts. The fifth and last paragraph relates to the content of the Implementing Act. Paragraphs two and four both relate to the composition of the scientific panel but are separated by paragraph three relating to its tasks. To provide a coherent analysis, this chapter is structured thematically rather than paragraph-by-paragraph.

5The following sub-section analyses the origins and development of the scientific panel in the legislative history.¹⁶ The next section, on substance, first addresses the purpose of the scientific panel and its tasks as laid down in Article 68(3).¹⁷ This is followed by sub-sections concerning the choice to exclude the scientific panel from the framework of so-called Commission expert groups,¹⁸ the composition and selection of members for the scientific panel,¹⁹ the particular roles of its members,²⁰ and lastly, the transparency of the scientific panel’s work.²¹

1.1. Legislative history

6The recognition of the need for a body of experts was present in the Commission’s ‘White Paper On Artificial Intelligence’ issued in early 2020.²² The paper envisioned a ‘committee of experts’ to provide assistance to the Commission, including in relation to identifying emerging trends and facilitating the implementation of the anticipated new legal framework.²³ However, once the Commission adopted the ‘Proposal for a Regulation on Artificial Intelligence’ in 2021 the proposal did not contain a ‘committee of experts’ nor any other equivalent to a scientific panel.²⁴

7The first indication of a scientific panel in the drafting process was introduced in the Council’s ‘General Approach’.²⁵ The General Approach advocated for a ‘central pool of independent experts’ which had the aim of supporting enforcement activities.²⁶ It is clear that this proposal influenced the role of the scientific panel in the AI Act because three of the four tasks in the proposal made it into the final text: supporting the work of MSAs based on request, supporting cross-border investigations, and supporting the Commission regarding Union safeguarding.²⁷ Thus, contrary to suggestions in earlier commentaries, the seeds of some of the functions of the scientific panel were planted by the Council’s General Approach.²⁸

8The European Parliament’s final negotiating position also included a proposal for a pool of experts.²⁹ However, the role of the pool of experts as envisaged by the European Parliament was less developed than the Council’s position. It was mentioned briefly in the proposed Article 56b(g), which suggested that the pool of experts would contribute to ‘developing the organizational and technical expertise required for the implementation’ of the proposed AI Act for Member States’ national supervisory authorities and the Commission.³⁰

9The orientation of the scientific panel towards GPAI models took shape during the trilogue along with the introduction of the concepts of GPAI models³¹, 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. ³² and qualified alerts.³³ In contrast to the Council’s General Approach, the ‘scientific panel of independent experts’ that came out of the trilogue negotiations is a standing advisory body (as opposed to a ‘pool of experts’ available on an ad hoc basis) with an express mandate to proactively issue ‘qualified alerts’³⁴ which can trigger follow-ups, such as investigations³⁵ and classification 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. ;³⁶ potential involvement during the drafting of the Codes of Practice;³⁷ the power to request 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 issue a request for documentation and information to 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;³⁸ and the potential for members of the scientific panel to be appointed to carry out evaluations on behalf of the Commission.³⁹ Thus, the scientific panel of independent experts in its current form was a result of the trilogue and it significantly differs from previous proposals in legal status, scope and powers. This orientation towards GPAI models and the ‘central role’ of the scientific panel as enshrined in the final text of the AI Act was also emphasised by the Commission press release announcing the political trilogue agreement.⁴⁰

10In summary, while an appreciation of external expertise was evident from the Commission’s White Paper on AI and some functions of the scientific panel were introduced with the Council’s General Approach, the central role of the scientific panel with regard to enforcement related to GPAI models as per Articles 68(3)(a), 90,⁴¹ 91(3) and 92(2) AI Act was only introduced in the trilogue negotiations.

2. Substance

2.1. Purpose and tasks of the scientific panel

11The purpose of the scientific panel is expressly to support the enforcement activities under the AI Act,⁴² with a particular focus on monitoring related to GPAI models, as emphasised by the recitals. Recital 113 highlights the role of qualified alerts from the scientific panel in complementing the GPAI monitoring activities 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. .⁴³ Recital 151 states that a scientific panel should be established to support the implementation and enforcement of the AI Act, ‘in particular the monitoring activities 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 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. ’.⁴⁴ Recital 163 states that the scientific panel should support the monitoring activities 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 complementing the governance systems for 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. ’.⁴⁵

12In accordance with the first sentence of Article 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. ’.⁴⁶ Providing more detail about the panel’s functions, Article 68(3) goes on to stipulate a non-exhaustive list of tasks that the panel shall ‘in particular’ carry out.⁴⁷ With this in mind, each task expressly listed in Article 68(3) is addressed in turn.

2.1.1. Support relating to GPAI models and systems (Article 68(3)(a))

13Article 68(3)(a) sets out five non-exhaustive⁴⁸ sub-tasks in which the scientific panel will ‘support[…] the implementation and enforcement of [the AI Act] 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’.⁴⁹ The first is ‘alerting 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. of possible 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 [GPAI models and systems]’ by issuing qualified alerts in accordance with Article 90.⁵⁰ This is a noteworthy mechanism in the enforcement activities 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. because such a request may trigger 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. , to exercise the powers laid down in Section 5 of Chapter IX AI Act⁵¹ as well as potentially trigger a classification of a GPAI model 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 Articles 51(1)(b) and 52(4) AI Act.⁵²

14The second sub-task is ‘contributing to the development of tools and methodologies for evaluating capabilities of [GPAI] models and systems, including through benchmarks’.⁵³ There is no express mention 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. . However, the mentioning of ‘benchmarks’ links the provision to the classification 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. because Recital 111, Article 51(1)(a), and Annex XIII all mention benchmarks in particular as a means to evaluate 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. for the purposes of classifying 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. .⁵⁴

15The third sub-task is advising on the classification 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. ,⁵⁵ and the fourth is advising on the classification of ‘various GPAI models and systems’.⁵⁶ Separation of the third and fourth sub-tasks suggests that the legislator intended to underscore the distinct nature 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. from GPAI models and GPAI systems, while emphasising that the scope of the scientific panel encompasses all three categories.⁵⁷ It is not clear what form such advice should take, but it could possibly include advice on when to update the compute threshold in Article 51 or how the Commission could exercise its discretion to make designation decisions.⁵⁸

16The fifth sub-task is contributing to the development of tools and templates.⁵⁹ ‘Tools’ are already mentioned earlier in the list in relation to evaluating capabilities.⁶⁰ In contrast, Article 68(3)(a)(v) refers to ‘tools and templates’ more generally, making clear that the contributions of the scientific panel need not be limited only to tools related to evaluations.

17In addition to this non-exhaustive list of sub-tasks, it can be contemplated what the role of the scientific panel may be in monitoring and advising 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. related to whether all 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. have been identified, assessed and mitigated to an acceptable level in compliance with Article 55(1) and the Code of Practice.⁶¹ For example, it might be within scope for the scientific panel to contribute to the identification of ‘state-of-the-art’ model evaluation techniques, risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. modelling methods, and Safety and Security Frameworks as referred to in the Safety and Security Chapter of the Code of Practice.⁶² Such monitoring and assessment would be in line with the broader purpose of the scientific panel elaborated above,⁶³ namely to support enforcement activities under the AI Act, and could arguably fall within the broad scope of Article 68(3)(a) AI Act.

2.1.2. Supporting the work of MSAs at their request (Article 68(3)(b))

18Article 68(3)(b) suggests that the scientific panel shall support the work of MSAs,⁶⁴ national authorities designated by Member States that have a different remit from 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. .⁶⁵ However, despite this distinction, the scientific panel’s support of MSAs’ work is listed under Article 68(3) under the heading of advising and supporting 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. . Article 69(1) states that ‘Member States may call upon experts of the scientific panel to support their enforcement activities under the [AI Act]’.⁶⁶ This sentence clarifies two important points. First, Member States can request support of the scientific panel. This is in line with the wording, ‘at their request’, used in Article 68(3)(b). Second, the scientific panel can be requested to support the enforcement activities of the Member States, indicated by the wording ‘their enforcement activities’. In light of this, one may ask whether it is coherent for Article 68(3)(b) to be included as a subparagraph under Article 68(3).

19The origin of this incoherence can be traced in the legislative history. As noted above, an expert body to support implementation of the AI Act was first introduced with the ‘pool of experts’ in the Council’s General Approach. This was before the introduction into the draft AI Act of a separate framework addressing GPAI models⁶⁷ and 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. , a separate body with exclusive powers to supervise and enforce the AI Act’s provisions on GPAI.⁶⁸ Thus, the ‘pool of experts’, as originally conceived, was designed to support the enforcement of the AI Act by national authorities, not a centralised body at EU level. Whilst the language ultimately incorporated in Article 68(3)(b)–(d) stayed largely unchanged from the Council’s proposal, subsequently 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 introduced into the governance framework of the AI Act with a specific remit over GPAI and the purpose and focus of the scientific panel was redirected to GPAI, with a corresponding set of new, weighty functions.⁶⁹

20It should be noted that, at the time of writing (September 2025), access to the experts by the Member States shall be facilitated by the Commission.⁷⁰ 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. shall provide for practical means for MSAs to submit requests and evaluate the necessity and proportionality of requests before either granting the request by appointing a rapporteur or refusing the request.⁷¹

2.1.3. Cross-border market surveillance activities (Article 68(3)(c))

21In accordance with Article 68(3)(c), the scientific panel may play a role in joint activities, for example when specific categories of high-risk 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. present a serious risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. across two or more Member States.⁷² Such activities naturally involve MSAs,⁷³ but they also involve 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 coordination role.⁷⁴ It is unclear from the wording of Article 68(3)(c) itself whether the panel is to support the MSAs directly or indirectly 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. .⁷⁵

2.1.4. Supporting 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 regard to the Union safeguard procedure (Article 68(3)(d))

22Article 68(3)(d) states that the scientific panel shall 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 carrying out duties in the context of evaluating whether measures taken by MSAs⁷⁶ are contrary to Union law in accordance with Article 81 AI Act.⁷⁷ Article 81 outlines the role of the Commission as evaluating and deciding whether national measures are justified, but it does not mention the scientific panel or 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. . Further, there is no mention in the Implementing Act of the Union safeguard procedure as laid out in Article 81.⁷⁸ Thus, at the time of writing (September 2025), the role of the scientific panel in such procedures remains unclear.

2.1.5. Prioritisation of tasks of the scientific panel

23The preceding sections demonstrate that the scientific panel has a broad range of tasks relating both 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 national MSAs.⁷⁹ However, the scientific panel does not have unlimited capacity. In practice, the potential tasks may exceed its capacity. This is expressly recognised in Article 14(3) Implementing Act, 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. shall consider the ‘available capacity of the scientific panel’ when evaluating requests by national MSAs for support from the scientific panel.⁸⁰ The work of the scientific panel will depend on the prioritisation of its different tasks. The question arises of who decides which tasks are prioritised? The role of the scientific panel itself, the Commission and the MSAs is discussed in turn below.

24The scientific panel is empowered to initiate several of its own tasks. In contrast to other expert groups, the scientific panel does not mainly operate ‘at the request’ or ‘by approval’ of the Commission.⁸¹ For example, in accordance with Article 7(2) Implementing Act, any panel member may initiate the preparation of qualified alerts.⁸² Such an alert may be issued if a simple majority of the scientific panel so approves.⁸³ Accordingly, the preparation and issuance of a qualified alert does not require prior request or approval from the AI Board, 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. or the Commission. The same holds true for other tasks, including requests for assistance from 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 obtain documentation or information 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 GPAI models⁸⁴ and requests for the Secretariat to organise thematic hearings.⁸⁵ The ability to initiate tasks without Commission request and/or approval requires the scientific panel itself to prioritise its tasks.

25On the other hand, the Commission can initiate tasks listed under Article 68(3) by appointing rapporteurs and contributors.⁸⁶ In accordance with Article 7(2) Implementing Act, the Commission, in consultation with the chair of the Scientific Panel, may appoint a rapporteur and two contributors for each of the tasks listed in Article 68(3).⁸⁷ The Commission is not required to ‘request’ or ‘approve’ activities of the panel, but these appointments will likely be decisive in the initiation and prioritisation of some of the scientific panel’s activities. Further, the Commission indirectly influences prioritisation by selecting the members of the scientific panel⁸⁸ and determining the content and timing of calls for expressions of interest from independent experts. The first call, for example, expressly sought expertise for implementation and enforcement of the AI Act as regards GPAI models and systems, ‘given the entry into application of the 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. ’.⁸⁹ This indicates the priority of tasks related to Article 68(3)(a) for the first two-year appointment of the scientific panel.

26Further, the MSAs can initiate tasks related to Article 68(3)(b) by requesting support from the scientific panel.⁹⁰ However, this process is facilitated 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. and requires 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 prioritise amongst such requests.⁹¹ Member States must substantiate the necessity and proportionality of requesting assistance,⁹² 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. will evaluate this, considering the ‘available capacity of the scientific panel and the necessity of ensuring effective access to experts for all Member States.’⁹³ 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 process a request within two weeks by either approving it and appointing a rapporteur and two contributors⁹⁴ or refusing the request giving reasons.⁹⁵ Thus, it is 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. to prioritise requests from MSAs in a manner that considers the capacity of the scientific panel (including the scientific panel’s capacity to initiate tasks itself related to GPAI models) as well as to ensure effective and timely access to expertise for Member States in line with Article 69(2) and (3) AI Act.⁹⁶

27To conclude, Article 68 sets out a broad scope of activities for the scientific panel. The provision expresses a clear focus on GPAI for the scientific panel, but its tasks also include support for MSAs and cross-border activities. The scientific panel, the Commission and the MSAs can all initiate tasks, but it is for the scientific panel (its members and the chair⁹⁷) and the Commission to prioritise between the tasks. Arguably, to fulfill the scientific panel’s monitoring purpose,⁹⁸ prioritisation of its tasks and the allocation of its resources should ensure that the scientific panel always has capacity to issue qualified alerts.

2.2. Nature of the scientific panel in light of the Commission expert group framework

28The scientific panel does not seem to have been modelled on any pre-existing expert panels or expert groups. In particular, it is relevant to note that the scientific panel does not constitute a so-called Commission expert group (“CEG”). CEGs are consultative bodies set up by the Commission to receive specialist advice from outside experts, providing a basis for sound policymaking.⁹⁹ For example, they advise the Commission in relation to the preparation of delegated acts, the implementation of EU legislation, and the preparation of implementing acts in fields such as climate policy, state aid, good manufacturing practices, and fertilisers.¹⁰⁰ As of 2025, there are more than a thousand active CEGs.¹⁰¹ While the scientific panel is an expert body that (amongst others) advises 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. , it is clearly not constituted within the framework for CEGs.¹⁰² This is in contrast to the high-level expert group on AI, a CEG set up in 2018 to advise the Commission on AI strategy;¹⁰³ the High-Level Group for the Digital Markets Act, a CEG established in 2023 to advise the Commission, on its request, on the implementation or enforcement of the Digital Markets Act;¹⁰⁴ and the AI Act Advisory Forum, a CEG set up in 2025 under the AI Act to provide technical expertise and advise the AI Board and the Commission.¹⁰⁵

29The choice to exclude the scientific panel from the CEG framework illuminates key characteristics of the scientific panel, giving some indication of the latter’s nature. At the outset, the choice may be explained by the fact that the scientific panel is not limited to advising the Commission in the form 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 also advises national MSAs on their request.¹⁰⁶ Thus, the tasks of the scientific panel exceed the typical remit of CEGs. However, the same applies to the Advisory Forum which does constitute a CEG.¹⁰⁷ More interestingly, there are (at least) five noteworthy differences between the scientific panel provisions and those typically applicable to CEGs. These differences provide important advantages for the scientific panel in fulfilling its advisory and supportive role in monitoring the fast-paced developments in the field of GPAI models.

30The first difference concerns operation. CEGs typically have to act on the request of the Commission or the chair acting with Commission approval.¹⁰⁸ In contrast, the scientific panel is set up to work more autonomously and act on its own initiative in several circumstances. For example, ‘any member’ of the scientific panel is allowed to initiate the preparation of a qualified alert.¹⁰⁹ This allows panel members to overcome potential procedural and informational bottlenecks.

31The second difference relates to tasks. The tasks of CEGs typically centre on the publication of opinions, recommendations and reports adopted by consensus.¹¹⁰ In contrast, the scientific panel is given powers in the AI Act itself to adopt qualified alerts¹¹¹ and request 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 support in requesting information 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. .¹¹² This requires the approval of a simple majority and a third of the members of the scientific panel, respectively.¹¹³ The scientific panel tasks are geared towards speedily alerting and informing the Commission in the context of an emerging and rapidly changing scientific field that presents great uncertainty and encompasses limited scientific consensus.

32The third difference concerns flexibility in rules of procedure. CEGs adopt rules of procedure mirroring standard rules of procedure.¹¹⁴ The standard rules may be departed from or supplemented when ‘justified by specific requirements’.¹¹⁵ In contrast, the scientific panel adopts its rules of procedure without any express requirement that these mirror standard rules of procedure. The requirements for the rules of procedure for the scientific panel laid out in the AI Act and the Implementing Act are more limited than standard rules of procedure are.¹¹⁶ This allows for more flexible, expedited procedures tailored to the unique tasks of the scientific panel and the rapidly evolving context in which it operates.

33The fourth difference pertains to remuneration. CEG members may only be remunerated in ‘exceptional cases’ on certain conditions,¹¹⁷ and such payment must not exceed 450 euros per full working day.¹¹⁸ In contrast, members of the scientific panel appointed as rapporteurs and contributors are to be remunerated for carrying out tasks requested 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. pursuant to Article 68(3).¹¹⁹ This may improve the ability of the scientific panel to attract technical talent in the context of fierce competition for scarce expertise in the field of GPAI models.

34The fifth difference relates to transparency. CEGs require publication of ‘all relevant documents’.¹²⁰ In contrast, the transparency obligations for the scientific panel are more limited.¹²¹ While it may, at times, be advantageous for the scientific panel to publish documents, the possibility for the scientific panel to effectively communicate confidential information 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 advise on sensitive matters presents its own advantages.

35Compared to the standard framework for CEGs, the scientific panel will, in theory, be able to operate more autonomously; raise speedy alerts rather than demonstrate scientific consensus; attract scarce talent through remuneration; adopt tailor-made rules of procedure allowing flexibility, speed and adaptation in a rapidly changing field; and balance trade-offs between transparency and handling sensitive information. These differences present particular advantages for the scientific panel in fulfilling its purpose and tasks. These are central characteristics that hint at the nature of the scientific panel and indicate that the legislator made an intentional, well-reasoned choice to establish the scientific panel outside the CEG framework.

2.3. Composition and selection of members

2.3.1. Overview

36The size of the scientific panel is omitted from the AI Act itself, but the Implementing Act clarifies that it shall consist of up to 60 members.¹²² The Commission, in consultation with the AI Board, sets out an exact number in each call for expression of interest.¹²³ The initial panel, which as of the time of writing (September 2025) has not yet been constituted, will consist of 60 members.¹²⁴

37Members are appointed for a term of two years with the possibility of the term being renewed.¹²⁵ The Commission may dismiss members if they resign or if they breach the obligations of independence, impartiality and objectivity; commitment; transparency; or confidentiality or the obligations of members or officials of the Union to not disclose information covered by the obligation of professional secrecy under Article 339 TFEU.¹²⁶

38The independent experts are to be selected on the basis of ‘up-to-date scientific or technical expertise in the field of AI necessary for the tasks of the scientific panel’.¹²⁷ Article 68(2)(a)–(c) and Article 68(4) AI Act set out four cumulative criteria that must be met for selection to the scientific panel. Furthermore, the overall composition of the panel must ensure fair gender and geographical representation.¹²⁸ These criteria are elaborated in the Implementing Act and in calls for expression of interest. They are addressed in turn below.

2.3.2. Expertise and competence in the field of AI (Article 68(2)(a))

39To be members of the scientific panel, experts must demonstrate ‘particular expertise and competence and scientific or technical expertise in the field of AI’.¹²⁹ The Implementing Act stipulates that expertise refers to ‘multidisciplinary and interdisciplinary adequate and up-to-date scientific, technical, or sociotechnical expertise related to AI, the impacts of AI or otherwise relevant to the effective enforcement of [the AI Act]’.¹³⁰ This includes ‘expertise pertaining to applied sectors, fundamental rights and equality, as appropriate’.¹³¹ In the June 2025 Commission call for expression of interest, the Commission has detailed eight areas that constitute up-to-date ‘relevant expertise’.¹³² In particular, the call includes two cumulative eligibility criteria regarding educational background and professional experience: (1) having obtained a PhD in a relevant area or equivalent experience and (2) having professional experience and proven scientific impact on AI/GPAI research or the study of AI impacts. It can be noted that the Implementing Act does not mention ‘competence’ a single time, compared to 13 mentions of ‘expertise’. Similarly, the call for expression of interest mentions ‘competence’ once (under the Privacy statement annexed) in contrast to 16 mentions of ‘expertise’. In this light, it is reasonable to interpret the requirement of ‘competence’ as satisfied by ‘expertise’.

2.3.3. Independence 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. (Article 68(2)(b))

40Experts who are appointed as members of the scientific panel shall demonstrate ‘independence from any 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 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. or 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. ’.¹³³ According to the Implementing Act, this requires that an expert sitting on the scientific panel is neither an employee nor in a contractual relationship with a 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 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. or GPAI models ‘which could affect their independence, impartiality and objectivity’ throughout the term of the panel appointment.¹³⁴ A literal interpretation of the quoted phrase suggests that contractual relationships with 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. may be permitted in principle if such relationships do not affect the independence, impartiality and objectivity of members. However, the first call for expression of interest states that ‘at the time of expressing interest to the scientific panel, and throughout the term of office, the candidate shall not be an employee of, or in a contractual relationship with a 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 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. or 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. ’.¹³⁵ This follows the more absolute wording of Article 68(2)(b) and leaves little room for a permissive interpretation of independence.

41The topic of independence in this context is not entirely straightforward. On the one hand, the importance of independence of the experts on the scientific panel is crucial, given the panel’s purpose of supporting enforcement activities under Article 68(1) AI Act and the panel’s tasks of advising and supporting 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 MSAs under Article 68(3). Furthermore, the importance of independent scientific expertise has increased over time in the EU institutional framework.¹³⁶ This is evidenced in the case law of the General Court, for example in the case of Alpharma Inc v Council, where the Court explicated that ‘expert scientific advice meeting the requirements of excellence, independence and transparency is of the utmost importance […] to ensure that the regulatory measures adopted by the Community institutions have a proper scientific basis and to ensure that the institutions were in a position to examine carefully and impartially all relevant evidence in a particular case’.¹³⁷ This broader significance supports a restrictive interpretation of the wording of Article 68(2)(b) discussed above.

42On the other hand, there is a tension between the strictness of the independence requirement and the number of experts (and their degree of expertise) who satisfy the criterion. Often, experts on a particular topic have conflicts of interest, for example through research funding, investments, consultation services, previous employment, etc.¹³⁸ This is particularly true in the field of GPAI models. In this field, academia dominated the development of notable models up until 2014, but its influence has rapidly decreased since then, with 90% of notable models developed by industry in 2024.¹³⁹ Valuable expertise rests with the individuals and organisations developing the most advanced models. Thus, there may be a very limited number of experts with a sufficient level of expertise regarding the most capable GPAI models and 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. who are not in a contractual relationship with a 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. . While this policy consideration does not bring to bear on the literal interpretation of the AI Act, it does hint at a potential conflict between a restrictive interpretation and the effective fulfillment of the purpose of the scientific panel.

43One potential way to alleviate this tension is a cooling-off period following the termination of a contractual relationship that compromised an expert’s independence, after which the expert could be appointed to the scientific panel.¹⁴⁰ The AI Act and the Implementing Act are silent on this matter. From a policy perspective, such a cooling-off period seems sensible in order for sufficiently independent expertise to join the scientific panel in the future, although the duration of such a period could jeopardise the currency of the individual’s particular expertise in the fast-paced evolution of GPAI science. It is likely to be necessary to assess such cooling-off periods on a case-by-case basis (rather than pursuant to an ex ante policy), given the variety of contractual relationships, the roles that candidate experts are likely to have had with 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. ,¹⁴¹ and the pace of change in the field.

2.3.4. Diligence, accuracy and objectivity when carrying out activities (Article 68(2)(c))

44Article 68(2)(c) stipulates that experts must demonstrate ‘an ability to carry out activities diligently, accurately and objectively’. This requirement is briefly elaborated on in the Implementing Act with some overlap with other requirements. Experts appointed to the scientific panel must commit to acting in the public interest and observe the principles of independence, impartiality, objectivity, commitment, transparency and confidentiality in accordance with Articles 10 to 13.¹⁴² Further, they must respond to requests from the Chair and the Secretariat and dedicate the necessary effort to complete assigned tasks to the best of their ability and within prescribed timelines.¹⁴³

2.3.5. Impartiality, objectivity, confidentiality, and instructions (Article 68(4))

45Article 68(4) makes four points which are further elaborated in the Implementing Act.¹⁴⁴ First, it states the requirement of impartiality and objectivity and stipulates that experts shall ensure the confidentiality of information and data obtained in carrying out their tasks and activities. Elaborating on this, Article 13 Implementing Act requires that experts appointed to the scientific panel sign a declaration of confidentiality, comply with Article 339 TFEU, and comply with Union classified information and sensitive non-classified information in accordance with Commission Decisions 2015/443 and 2015/444.¹⁴⁵ Second, Article 68(4) stipulates that experts shall ‘neither seek nor take instructions from anyone when exercising their tasks’. Relatedly, the Implementing Act states that experts are appointed in their personal capacity and cannot delegate their responsibilities to any other person.¹⁴⁶ Third, Article 68(4) requires that experts make a declaration of interests to be published, as is standard under EU administrative law.¹⁴⁷ The Implementing Act elaborates that this declaration must indicate any interest that may compromise or reasonably be perceived to compromise their independence, impartiality and objectivity, including in respect of close family members.¹⁴⁸ In the event of a delay before appointment, this declaration shall be updated before appointment to the panel (or reserve list)¹⁴⁹ and when changes of circumstances occur.¹⁵⁰ Moreover, experts must sign a declaration of commitment to public interest and the principles of independence, impartiality, objectivity, transparency and confidentiality.¹⁵¹ Fourth, Article 68(4) requires 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 establish ‘systems and procedures to actively manage and prevent potential conflicts of interest’.

46Requirements of impartiality, objectivity and confidentiality are important in panel appointees given the nature of the tasks that members of the scientific panel may perform, not least making qualified alerts under Article 90 AI Act and conducting evaluations under Article 92(2) AI Act. As pointed out by scholars, the main source of legitimacy for expert bodies, like the scientific panel, is the objectivity and quality of their output.¹⁵² Further, Article 41 of The Charter of Fundamental Rights enshrines the right to have affairs handled impartially by EU institutions, bodies and agencies.¹⁵³ This right is well-established in case law, and the General Court has held that the requirement of impartiality extends to experts consulted in the context of the adoption of Commission decisions.¹⁵⁴ This is relevant because certain actions of the scientific panel may affect Commission decisions. For example, the Commission may adopt a decision designating a GPAI model as presenting 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. following a qualified alert by the scientific panel in accordance with Article 52(4) AI Act.¹⁵⁵

2.3.6. Fair gender representation (Article 68(2))

47The second paragraph of Article 68(2) states that the Commission shall ensure ‘fair gender […] representation’. The Implementing Act adds the qualifier ‘to the extent possible’, suggesting that there is no set gender quota or absolute quantitative requirement.¹⁵⁶ The Implementing Act states that the Commission must give preference to the under-represented gender when deciding between two equally qualified candidates,¹⁵⁷ but is otherwise silent on the topic of gender balance.¹⁵⁸ This is in contrast to the requirement for ‘fair geographical representation’, where the Implementing Act lays down quantifiable conditions.¹⁵⁹

2.3.7. Fair geographical representation (Article 68(2))

48According to the second paragraph of Article 68(2), the Commission shall ensure ‘fair […] geographical representation’. The Implementing Act gives more detail, stating that the Commission shall appoint at least one expert, and up to a maximum of three experts, from each Member State as well as each member of the EFTA/EEA countries on two conditions: (1) there are applicants from that country who satisfy the criteria stipulated in the relevant call and (2) a sufficiently comprehensive coverage of relevant areas of expertise can be achieved. Thus, representation from every Member State is not an absolute criterion. However, the Implementing Act sets out an absolute criterion that Member State nationals and EFTA/EEA nationals shall make up at least 80% of the selected experts, allowing up to 20% of the appointed experts from third countries.¹⁶⁰

2.4. Particular roles for members of the scientific panel under the Implementing Act

49While the AI Act is silent on the organisation of the scientific panel, the Implementing Act makes clear that members of the scientific panel can be appointed for particular roles: chair, vice-chair, rapporteurs and contributors. The chair and vice-chair mainly have organisational responsibilities. One rapporteur and two contributors can be appointed per task performed by the scientific panel. Certain actions are reserved exclusively for rapporteurs, namely requests for assistance made 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 accordance with Article 91(3) AI Act. Lastly, most actions may be initiated by any member, including preparing qualified alerts. Each role is discussed in turn below, followed by a short note on rules of procedure.

2.4.1. Chair and vice-chair

50Two members will be selected as chair and vice-chair by the Commission upon majority recommendation by the members of the scientific panel.¹⁶¹ The term of office for these positions lasts for the term of the scientific panel, that is two years, and is renewable once.¹⁶² The chair’s responsibilities as outlined in the Implementing Act are mainly organisational. The chair is to consult the Commission on the latter’s appointment of members as rapporteurs and contributors in different situations, in particular the appointment of members to perform tasks of the scientific panel listed under Article 68(3) AI Act¹⁶³ and the appointment of another member when an expert is no longer able to effectively perform their allocated task.¹⁶⁴ Additionally, when 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. decides to launch a measure pursuant to Articles 91 to 93 AI Act, Article 19(4) Implementing Act 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 appoint, in consultation with the chair, a rapporteur and two contributors to advise it.¹⁶⁵ As Article 19(4) Implementing Act does not expressly mention qualified alerts, it is unclear from the provision itself whether this appointment and consultation requirement only applies when 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. adopts measures pursuant to Articles 91 to 93 AI Act as a response to a qualified alert, or whether it applies whenever 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. adopts measures pursuant to Articles 91 to 93 AI Act. However, from a systematic view, Article 19 Implementing Act is titled ‘Handling of qualified alerts’. Article 19(1) Implementing Act requires 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 evaluate the qualified alert; to decide which, if any, measures to take; and to close the alert if it decides not to launch any measures following a qualified alert. Articles 19(2) and (3) Implementing Act lay down 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 evaluation process of qualified alerts, including giving 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 right to request more information from the scientific panel to facilitate 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 decision-making. In these circumstances, it is reasonable to conclude that Article 19(4) Implementing Act should be read as outlining the implications when 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. decides to launch measures in response to a qualified alert. Lastly, the chair is also involved when members want to conduct a thematic hearing with stakeholders to gather evidence, as it is the chair that shall request the Secretariat to organise such a hearing following the request of at least three members of the scientific panel.¹⁶⁶

2.4.2. Rapporteurs and contributors

51In accordance with the first sentence of Article 7(2), the Commission, in consultation with the chair, may appoint a rapporteur and two contributors ‘for each task pursuant to Articles [sic] 68(3)’.¹⁶⁷ The purpose of this organisational structure is to ‘ensure flexibility so that specialised knowledge can be deployed based on requisite needs.’¹⁶⁸ When appointing rapporteurs, the Commission and chair should take into account the candidates’ expertise, availability, possible conflicts of interest, and security concerns.¹⁶⁹ Appointed rapporteurs and contributors will be remunerated when carrying out tasks for the scientific panel.¹⁷⁰ The roles of rapporteurs and contributors are significant for (at least) two reasons.

52First, according to the Implementing Act, the Commission is required to appoint a rapporteur in two distinct situations. The first situation is when an MSA¹⁷¹ requests support from the scientific panel 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. concludes that the request is necessary and proportionate.¹⁷² Then, the Commission shall appoint a rapporteur and two contributors. The second situation relates to Article 19(4) Implementing Act.¹⁷³ This Article stipulates that the Commission is required to appoint a rapporteur and two contributors from the scientific panel when, following a qualified alert, 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. launches a measure pursuant to Articles 91 to 93 AI Act in order to receive advice about the adoption of the measures.¹⁷⁴

53Second, according to the Implementing Act, certain panel activities may only be initiated by rapporteurs. Namely, requests for assistance made 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 accordance with Article 91(3) AI Act may only be issued by an appointed rapporteur with the support of at least one-third of the panel members.¹⁷⁵ If such a request is granted, the information received shall only be made available to the rapporteur and contributors appointed for the specific task for which the assistance was requested.¹⁷⁶

2.4.3. Members

54In accordance with the second sentence of Article 7(2) Implementing Act, any panel member may initiate the preparation of qualified alerts and perform other tasks of the scientific panel. This sentence stipulates that ‘members of the scientific panel may at any time decide to prepare qualified alerts pursuant to Article 90 [AI Act] or other tasks of the scientific panel on their own initiative’.¹⁷⁷ The text refers to ‘members of the scientific panel’ rather than ‘rapporteurs’, ‘the Chair’, or ‘appointed members’, suggesting that any member may take such an initiative. On a systematic interpretation, the second sentence of Article 7(2) clarifies that its first sentence is not to be understood as generally requiring Commission initiative or the appointment of rapporteurs for the initiation of panel tasks. Unless otherwise specified, members can take their own initiative to perform the tasks of the panel.¹⁷⁸

55From a policy perspective, this possibility for members to take actions of their own motion indicated by the second sentence of Article 7(2) may prevent undue delays, for example from bottlenecks if the Commission is slow to or fails to appoint rapporteurs. It does confer a degree of agency on all members of the panel, ensuring that the chair or rapporteurs do not have de facto veto powers over the preparation of qualified alerts or other tasks.¹⁷⁹ It could be argued that this organisational structure could bring with it the unilateralist’s curse, where a number of experts could take an action with far-reaching consequences without requiring the permission of others, and thereby leading to more frequent or less co-ordinated action than is optimal.¹⁸⁰ However, this risk Article 3(2) AI Act: ‘risk’ means the combination of the probability of an occurrence of harm and the severity of that harm. seems to be mitigated because a simple majority of members is required for raising qualified alerts under Article 90 AI Act¹⁸¹ and only rapporteurs, with the support of one-third of the scientific panel, are empowered to request assistance from 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 obtain information from a 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 91(3) AI Act.¹⁸²

2.4.4. Rules of procedure

56The rules of procedure of the scientific panel may provide further clarification on the responsibilities of different roles outlined above. Once the scientific panel has been constituted, the members will adopt rules of procedure by a simple majority.¹⁸³ The Secretariat, a role jointly performed 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. and the Joint Research Centre,¹⁸⁴ is tasked with monitoring compliance with the rules of procedure.¹⁸⁵ The scientific panel may have some flexibility in determining the design and content of the rules of procedure as the panel does not constitute a so-called Commission expert group.¹⁸⁶ However, the Implementing Act stipulates that the rules of procedure shall at least provide the procedure for carrying out the tasks of the scientific panel referred to in Article 68(3);¹⁸⁷ rules to ensure the application of the principles of independence, impartiality, objectivity, commitment, transparency and confidentiality in accordance with Articles 10 to 13;¹⁸⁸ and rules related to voting, including through silence procedures.¹⁸⁹ Furthermore, the Implementing Act explicitly allows for more specific procedures relating to issuing qualified alerts in accordance with Article 90(1) to be introduced in the rules of procedure.¹⁹⁰ The rules of procedure are to be made public, which supports the transparency of the panel’s work addressed below.¹⁹¹ Further, they shall be reviewed, and where necessary updated, every two years by the scientific panel in agreement with the Secretariat.¹⁹²

2.5. Transparency

57Article 68 AI Act is largely silent about particular requirements regarding transparency of the scientific panel’s work, with the exception of Article 68(4) which requires that experts draw up a public declaration of interest. However, Article 12 Implementing Act states that the activities of the panel shall be ‘carried out in a transparent manner’.¹⁹³ This has the important function of ensuring trust in the work of the scientific panel.¹⁹⁴ A dedicated Commission website will contain the list of experts along with their CVs and declarations of interest, confidentiality and commitment.¹⁹⁵ As noted above, the rules of procedure of the scientific panel are to be made public.¹⁹⁶ Further, the opinions and recommendations that the panel provides when fulfilling tasks listed in Article 68(3) shall be public if they do not disclose confidential business information, trade secrets, or strategic interests of the Union.¹⁹⁷ The participation and conclusions from thematic hearings shall also be public.¹⁹⁸ These requirements for transparency are at the centre for the delicate balancing, case-by-case, between transparency of the scientific panel’s work on the one hand and confidentiality and business secrecy on the other hand.