Cambridge Commentary on EU General-Purpose AI Law

1Prior to the exercise of certain enforcement powers against 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. , the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 initiate a ‘structured dialogue’ with said 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. . Structured dialogues are provided for in three separate articles of the AI Act:

1. Article 91(2) AI 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. may initiate a structured dialogue with 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 a GPAI model before sending a request for information.²
2. Article 92(7) AI 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. may initiate a structured dialogue ‘to gather more information on the internal testing of the model, internal safeguards for preventing 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 other internal procedures and measures 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. has taken to mitigate 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. ’ prior to requesting access to a GPAI model.³
3. Article 93(2) AI 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. may initiate a structured dialogue 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 a GPAI model before a measure is requested.⁴ Article 93(3) AI Act elaborates that if 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 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. offers commitments to implement mitigation measures during a structured dialogue, the Commission may make those commitments binding by a decision and declare no further grounds for action.⁵

2As already noted, there is no definition of a structured dialogue in the AI Act, and even the origin of the term ‘structured dialogue’ in the context of the AI Act remains unclear. The AI Act does not describe what a structured dialogue entails nor does it establish a uniform process by which a structured dialogue is to be carried out. Only the structured dialogue provision in Article 92(7) identifies a purpose for that process: to gather more information on internal testing of the model, internal safeguards for preventing 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. , and other internal procedures and measures 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. has taken to mitigate 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. .⁶ In contrast, Articles 91(2), 93(2) and 93(3) do not specify a purpose for structured dialogues in those contexts. That said, the terms of Article 93(3) perhaps allow the purpose of structured dialogues under Article 93(2) to be inferred from the reference to 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. ’s offer of a potentially binding commitment. At the time of writing (March 2026), there is no specific Commission guidance on the procedure or conduct of structured dialogues under Articles 91 to 93 AI Act.

3In the absence of guidance as to the character and content of structured dialogues in the AI Act, it is instructive to note that references to ‘structured dialogues’ can be found in other EU legislative acts. These other examples of structured dialogue do not relate to regulatory enforcement but to cooperation mechanisms between Member States and the Commission,⁷ inter-institutional dialogues⁸ or dialogues between bodies at the Union level and various private or non-state actors (civil society, industry, etc.). What follows are brief discussions of four illustrative examples.

4Under Article 13(1)(m) European Media Freedom Act (“EMFA”), the European Board for Media Services, which is composed of national authorities, is tasked with regularly organising a ‘structured dialogue’ between 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 very large online platforms, representatives of media service 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 civil society in accordance with Article 19 EMFA.⁹ The purpose of these structured dialogues is to discuss experiences and best practices as regards specific provisions under the EMFA, foster access to diverse offerings of independent media on very large online platforms, and monitor adherence to self-regulatory initiatives.¹⁰ The results of such dialogues are to be reported to the Commission and, where possible, made publicly available.¹¹ This use of structured dialogues is quite distinct from structured dialogues under the AI Act, both with regard to the parties involved (multi-stakeholder dialogue organised by a board of national authorities in contrast to bilateral dialogue initiated by the Commission) and the purpose of dialogues (discussing experiences and best practices, fostering access and monitoring adherence to self-regulatory initiatives with public reporting where possible, as opposed to an element in investigation of compliance and/or enforcement in cases of non-compliance). Structured dialogues themselves are not directly relevant to enforcement powers in the EMFA, although recital 56 EMFA does state that the Commission could examine the reports of the results of such structured dialogues when assessing systemic and emerging issues across the Union as part of its enforcement of the Digital Services Act (“DSA”).¹² 

5Turning to the DSA itself, Article 36 refers to ‘dialogue’ between the Commission and 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. in the context of a crisis response.¹³ This dialogue takes place after the Commission has adopted a decision requiring 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. to take certain actions. In this context, the dialogues are a mechanism to discuss the effectiveness and proportionality of measures intended or implemented by 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 whether these measures meet certain requirements. This is in contrast to structured dialogues under the AI Act, which take place prior to the adoption of Commission decisions under Chapter IX AI Act.

6The Digital Markets Act (“DMA”) does not use the term ‘structured dialogue’. However, under Article 8(3) DMA, there is a process whereby a designated gatekeeper may submit a reasoned request asking the Commission to assess whether the measures it has implemented or plans to implement comply with the relevant provisions of the DMA.¹⁴ Thus, Article 8(3) establishes the right of a regulated entity to seek a dialogue with the Commission to determine whether chosen measures comply with the relevant provisions of the DMA. There are features that distinguish a dialogue under the DMA from a structured dialogue under the AI Act. A dialogue under the DMA is a mechanism that provides a channel for gatekeepers to seek regulatory clarity, whereas under the AI Act, a structured dialogue is a discretionary and preliminary investigative or enforcement step that may be initiated 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. . That said, both types of dialogue are ex ante and may, in certain circumstances, lead to a more cooperative investigative or enforcement relationship between the regulated entity and the regulator.

7Structured dialogues are also mentioned in Decision (EU) 2022/2481 establishing the Digital Decade Policy Programme 2030 as a cooperation mechanism between Member States and the Commission. Article 8(7) of Decision (EU) 2022/2481 concerning the deviation in national projected trajectories describes in detail the basis, aim and intended result of the structured dialogue.¹⁵ This stands in stark contrast to the lack of detail regarding structured dialogues in the AI Act. Only Article 93(3) AI Act envisages a particular possible outcome from a structured dialogue, namely, that 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. may offer commitments which the Commission may make binding on 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. .¹⁶ 

8These four examples from other pieces of EU legislation suggest that other ‘dialogue’ or ‘structured dialogue’ mechanisms provide limited insight into the structure and character of structured dialogues under the AI Act.¹⁷ It has been suggested by some commentators that structured dialogues are inspired by the practice of dialogue or conciliation in sanction procedures (in German ‘Verständigungsgespräche’), which, according to those authors, is a common practice of the German Federal Network Agency, the Federal Financial Supervisory Authority, and data protection authorities.¹⁸ Nonetheless, it seems premature at this stage, and without further guidance, to articulate with certainty the intention and inspiration of the European legislature for structured dialogues in the AI Act.

9Without adopting an overly literal approach, the use of the word ‘structured’ may be of assistance in gaining a better understanding of the ‘structured dialogue’ mechanism under the AI Act. For the dialogues to be structured, it might be expected that they would follow a certain procedure or form. For example, there may be certain steps, stages or timeframes of such a procedure that would be preordained, and requirements or expectations as to whether (and when) such procedures are to be written, oral or in person. The same commentators cited above have suggested that the procedure and form are to be determined 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. .¹⁹ It is unclear to what extent a similar procedure and form must be followed for different instances of structured dialogues under each of Articles 91 to 93 AI Act, or whether a dialogue is ‘structured’ as long as there is a structured procedure and form laid out in each particular instance of a dialogue. Even if structured dialogues are intended to be less formal than the adoption of formal measures against 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. , exactly how informal and cooperative they will be remains to be determined.

10While much is unclear in the context of investigation and enforcement of EU law obligations, at the level of principle it is clear that dialogue can be used as a tool to promote compliance before adopting coercive measures. In theory, it is conceivable that under the AI Act structured dialogues may provide a process of communication and interaction between the regulator and the regulated entity that benefits both parties, including faster processes, greater compliance, more valuable exchange of information, and resource savings. This is reflected in the Guidelines on the scope of the obligations for GPAI models, which 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. will take a collaborative, staged, and proportionate approach’ 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. expects 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. to collaborate and actively engage 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. at the later stages of potential formal proceedings conducted in line 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. ’s powers under the AI Act (for example, through structured dialogues)’.²⁰ Structured dialogues also allow for agreement between the Commission and the 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. to be reached without the need for a Commission decision, which could potentially be challenged under Article 263 TFEU.²¹ A similar practice is common in EU competition proceedings where agreement is reached between the Commission and undertakings under investigation, following which the case is concluded without formal findings being made.²² While Article 93(3) AI Act envisages an outcome from the structured dialogue that can be made binding by way of Commission decision, there is nothing to prevent agreements being reached in the course of structured dialogues initiated under Article 91 or Article 92.²³ It is of note that a breach of binding commitments is not expressly included as a ground for issuing a fine in Article 101 AI Act, while it is under Article 74(1)(c) DSA and Article 30(1)(e) DMA.²⁴ It is important to note that commitments made under Article 93(3) are, however, of relevance to fines imposed under Article 101 AI Act. Article 101(1) AI Act provides that the Commission, when fixing the quantum of the fine or periodic penalty, shall, among other things, take into account commitments made under Article 93(3).²⁵ 

11It is clear from the wording of Articles 91(2), 92(7) and 93(2) AI Act (‘may initiate a structured dialogue’) 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 discretion over whether to initiate structured dialogues and cannot be bound to do so 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. .²⁶ The Guidelines on the scope of obligations for GPAI Models emphasise a ‘collaborative and staged’ approach to supervision, investigation, enforcement and monitoring under the AI Act, and, in light of that emphasis, it is likely that structured dialogues will indeed play their part in the enforcement of the AI Act.²⁷ However, there is no provision of the AI Act that requires the Commission and/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. to utilise the structured dialogue procedure prior to exercising their enforcement powers. This could be important in cases of urgency should the Commission determine that faster action is required.²⁸

12An illustrative analogy can be drawn from EU competition law, wherein provision is made for both a simple request and a request for information by decision of the Commission.²⁹ In that context, it is well established that requests for information made by the Commission to an undertaking must comply with the principle of proportionality: the obligation imposed on an undertaking to supply information should not be a burden on the undertaking that is disproportionate to the needs of the inquiry.³⁰ However, the Court has held that the Commission could not be considered to have infringed the principle of proportionality by ‘merely’ having adopted a request for information by a decision without first adopting a simple request in light of the circumstances of the case.³¹ There is a comparison between the simple request procedure in EU competition law and structured dialogues: both are non-mandatory processes by which enforcement aims can be achieved. Thus, the decision to initiate a structured dialogue remains at the discretion 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. , having regard to the principle of proportionality.

13As a matter of practice, initiating a structured dialogue under the AI Act would be consistent with the broader approach followed by the Commission in the enforcement of other areas of EU law to attempt to resolve matters by way of exchanges prior to the adoption of formal binding decisions. As noted above, it mirrors enforcement structures under EU competition law³² and the DSA³³ by which the Commission can request information either by ‘simple request’ or by decision. In contrast to a decision, a simple request is not binding on the regulated entity, which is not forced to answer the request, although it may face consequences if it chooses not to do so. However, if the regulated entity does decide to answer a simple request, it cannot provide incorrect or misleading information as part of that answer.³⁴ The extent to which structured dialogues can be considered analogous to ‘simple requests’ is unclear from the law as it stands. One readily apparent important difference, however, is that both Regulation 1/2003 and the DSA provide for minimum formal requirements for simple requests,³⁵ whereas the AI Act is silent on formal requirements for structured dialogues.

14Although structured dialogues are not mandatory under the AI Act, they could be an important enforcement tool available 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. , particularly to encourage cooperation 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. . As noted above, the Guidelines 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. ‘will take a collaborative, staged, and proportionate approach’ to enforcement of GPAI provisions.³⁶ For early stages, the Guidelines mention ‘close informal cooperation 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. during the training of their GPAI models in order to facilitate compliance and ensure timely market placement’.³⁷ Further, ‘the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. expects 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. to collaborate and actively engage 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. at the later stages of potential formal proceedings […] (for example, through structured dialogues)’.³⁸ As observed above, the limits of how cooperative or collaborative in practice such dialogues can, or must, be will likely need to be determined by the Court of Justice of the European Union (“CJEU”). In any event, it is worth noting that dialogues remain asymmetrical in the sense that this procedure involves a regulator and a regulated entity; it is not an exchange between two equal actors,³⁹ and the relative power of each actor is likely to shift dynamically from one phase of the dialogue to another. For example, as a structured dialogue moves towards a conclusion it may be in the interests of the 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. to seek a resolution within the structured dialogue in order for 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. to shape the outcome in a way that is more favourable than a Commission decision – to which the only recourse is to challenge its legality pursuant to Article 263 TFEU – might be.

15A further area of uncertainty relates to which duties exist 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. in scenarios where the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 decide to initiate a structured dialogue. The AI Act does not have explicit grounds for fines to be issued 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. who fail to respond to, or decline, invitations 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 engage in a structured dialogue.⁴⁰ This may suggest that requests formulated 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. as part of structured dialogues are not legally binding. In a similar vein, it has been argued in the literature that 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. is not compelled to answer an invitation 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. or continue the dialogue.⁴¹ However, other considerations suggest that the consequences of being invited to participate in a structured dialogue are more nuanced. In particular, the duty to cooperate set out in Article 53(3) AI Act may require 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. to cooperate in relation to structured dialogues. Article 101(1)(b) AI Act may require 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. not to provide false information during structured dialogues. At the time of writing (March 2026), much remains unclear about the legal effects 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. using its discretion to initiate a structured dialogue. In response to this uncertainty, this section discusses two different interpretations⁴² and the consequences for a duty to cooperate 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 in relation to structured dialogues.

16The starting point of this analysis is Article 53(3) AI Act which explicitly sets out a general duty to cooperate 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.⁴³ It states that ‘[p]roviders 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. shall cooperate as necessary with the Commission and the 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. in the exercise of their competences and powers pursuant to this Regulation.’ This wording suggests that the material scope of the duty to cooperate in Article 53(3) is broad, covering the AI Act in its entirety (‘in the exercise of their competences and powers pursuant to this Regulation’).⁴⁴ Additionally, recital (e) of the Safety and Security Chapter of the adopted GPAI Code of Practice refers to the principle of cooperation laid down in Article 53(3) AI Act.⁴⁵ While the Code of Practice is not binding upon 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. , the inclusion of recital (e) emphasises the importance of the duty of cooperation in relation to 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. . Further, the fact that Article 53(3) requires cooperation with the Commission, whereas the discretion to commence a structured dialogue rests 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. , is addressed and resolved by Article 3(47) AI Act, which provides that ‘[…]references in this Regulation 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’.⁴⁶

17The key question is 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. ’s initiating a structured dialogue with a 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. constitutes the exercise of a ‘power’ or ‘competence’ within the meaning of Article 53(3) AI Act. The answer to this question presents some difficulty. On the one hand, the wording ‘the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 initiate a structured dialogue’ (emphasis added) indicates a conferral of a power 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. to initiate structured dialogues. If the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 discretion to commence a structured dialogue is indeed a power, there is arguably no reason it should not be covered by the duty of cooperation articulated in Article 53(3). In that case, failure to fulfil the duty to cooperate may lead to fines for infringement of Article 53(3) under Article 101(1)(a) AI Act, and one must carefully analyse the extent of the duty to ‘cooperate as necessary’ in any given instance. On the other hand, the AI Act does not expressly state that a structured dialogue is a mandatory process. Nor does Article 101 provide for fines explicitly in relation to structured dialogues. This may suggest that the discretion to initiate structured dialogues is not a power. In this case, there may be, as a minimum, a good faith obligation applying to 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. to not provide misleading or false information if 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. does engage with a structured dialogue. The subsections below examine these two different interpretations.

18Before turning to those two interpretations, it is worth noting two arguments addressing the general applicability of Article 53(3) AI Act. First, there may be an argument that the reference in Article 53(3) to ‘cooperate as necessary’ is too vague to impose concrete 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. as it offends the principle of legal certainty.⁴⁷ Some provisions in the AI Act conferring duties to cooperate appear more detailed than Article 53(3). For example, Article 54(3)(d) AI Act states that authorised representatives Article 3(5) AI Act: ‘authorised representative’ means a natural or legal person located or established in the Union who has received and accepted a written mandate from a provider of an AI system or a general-purpose AI model to, respectively, perform and carry out on its behalf the obligations and procedures established by this Regulation. 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 GPAI models shall be empowered to ‘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. and competent authorities, upon a reasoned request, in any action they take in relation to 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. ’ (emphasis added).⁴⁸ However, other provisions in the AI Act posit duties to cooperate in similarly vague terms as Article 53(3). Articles 23(7), 24(6) and 26(12) AI Act specify that importers Article 3(6) AI Act: ‘importer’ means a natural or legal person located or established in the Union that places on the market an AI system that bears the name or trademark of a natural or legal person established in a third country. , distributors Article 3(7) AI Act: ‘distributor’ means a natural or legal person in the supply chain, other than the provider or the importer, that makes an AI system available on the Union market. and 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. respectively shall ‘cooperate with the relevant competent authorities in any action those authorities take in relation to a high-risk 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.  […]’ (emphasis added). Taking a broader perspective, it is not uncommon for provisions in EU digital law to entail somewhat vague duties to cooperate. For example, Article 31 GDPR states: ‘The controller and the processor and, where applicable, their representatives, shall cooperate, on request, with the supervisory authority in the performance of its tasks.’ (emphasis added).⁴⁹ Further, Article 13(2) DSA, Article 28(5)(d) DMA, and Article 23(4) Corporate Sustainability Due Diligence Directive contain similarly vague duties to cooperate.⁵⁰ It is appropriate to acknowledge that none of these provisions has been subjected to the CJEU’s scrutiny of their lack of specificity. Nevertheless, while legal certainty may require a confined reading of Article 53(3) AI Act,⁵¹ the provision does not stand out as being detrimentally vague when compared to similar provisions in the AI Act and in other EU digital laws and cannot be disregarded altogether as incapable of giving rise to concrete obligations.

19Second, it could be argued that provisions requiring 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. to cooperate in specific instances, namely to provide requested information in Articles 91(5) and 92(5) AI Act, suggest that the legislature did not intend for Article 53(3) to be generally applicable. Articles 91(5) and 92(5) AI Act state that 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. or their representative ‘shall supply the information requested’ (emphasis added). It could be argued that if Article 53(3) AI Act sufficiently established a duty to cooperate in relation to the powers in Article 91 to 93 AI Act, the inclusion of the obligation to supply information in Articles 91(5) and 92(5) AI Act would be redundant. However, the inclusion of these specific duties arguably serves to particularise the existence of the duty to cooperate laid down in Article 53(3) with regard to the provision of information requested rather than to indicate that there is no general duty to cooperate in the exercise of the competences and powers of the Commission. The opposite interpretation would have the unattractive implication that 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. need not cooperate with the Commission in response to requests made pursuant to Article 93 AI Act, as this Article does not contain any explicit duty to cooperate. Thus, the more attractive, systematic interpretation of Articles 53 and 91 to 93, when read together, is that the more specific obligations to cooperate in Articles 91(5) and 92(5) AI Act do not undermine the existence of a generally applicable duty 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. to cooperate with the Commission 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. under Article 53(3) in the exercise of their powers and competences pursuant to the AI Act. On this interpretation, Article 53(3) AI Act imposes a context-specific, and therefore dynamic, duty of cooperation on 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. , requiring it to take action to cooperate according to the particular circumstances and the demands of the general principles of EU law, such as the principle of proportionality.⁵²

3.1. Interpretation 1: Structured dialogues are a power for the purpose of Article 53(3) AI Act

20The fact that structured dialogues are initiated 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. under Articles 91(2), 92(7) and 93(2) AI Act 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. has been given what one might neutrally call a function in relation to structured dialogues. The question is 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. ’s function can be considered a ‘power’ or ‘competence’ for the purposes of Article 53(3). Neither ‘powers’ nor ‘competences’ are defined terms under the AI Act. However, it is well established that the terms ‘competence’ or ‘competences’ are usually understood in EU law as referring to the areas of responsibility of the Union on the one hand and the Member States on the other, in accordance with Article 2 TFEU.⁵³ This is reflected in Article 53(3) AI Act itself and in Article 88(1) AI Act,⁵⁴ such that it seems reasonable to proceed on the basis that this orthodox meaning of ‘competences’ was intended. As for the interpretation of ‘powers’, regrettably the Article 53(3) duty to cooperate is not explained or clarified in the recitals, leaving its interpretation largely open.⁵⁵ The following analysis focuses on whether the commencement of a structured dialogue can be considered a ‘power’, based on a systematic and/or teleological interpretation of Articles 91(2), 92(7) and 93(2) AI Act, given their location not only in Section 5., 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. ’, but also within Articles 91, 92 and 93 AI Act, which are all titled as ‘powers’.

21There are convincing arguments that the discretion to initiate a structured dialogue constitutes a power. On one view, structured dialogues prima facie constitute powers based on the use of the wording ‘the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 initiate a structured dialogue’. The AI Act contains numerous provisions that confer powers using the term ‘may’, followed by some specified action. Articles 91 to 93 AI Act, within each of which the discretion to commence a structured dialogue is located, are key examples. To illustrate this by considering one of those provisions, Article 92 is titled ‘Power to conduct evaluations’. Article 92(1) confers 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. the power to conduct evaluations, stating 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. , after consulting the Board, may conduct evaluations’ of GPAI models to assess compliance and investigate 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. at Union level. Next, according to Article 92(2), ‘[t]he Commission may decide to appoint independent experts to carry out evaluations on its behalf, including from the scientific panel’. Under Article 92(3), ‘the Commission may request access to 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 through APIs’. Thus, Articles 92(2) and (3) contain related, additional powers conferred on the Commission to support the exercise of the main ‘power to conduct evaluations’ to which Article 92 is primarily addressed. In this way, Article 92 AI Act shows that supplementary powers have been included by the legislature in the enforcement provisions of the AI Act to ensure that the main enforcement power, named in the provision’s title, is optimised to achieve the most suitable enforcement outcome in the circumstances.

22This interpretation of the initiation of structured dialogues as a supplementary power entrusted 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. to assist the Commission’s enforcement work is supported by Article 88(1) AI Act, which requires the Commission to ‘entrust the implementation of [its exclusive powers to supervise and enforce Chapter V 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. ’. It should also be noted that there is no part of Article 53(3) or Section 5. Chapter IX AI Act that explicitly indicates that the structured dialogues are not a power and/or that the duty of cooperation is inapplicable to structured dialogues. Accordingly, there seems to be a compelling argument that the commencement of structured dialogues constitutes a supplementary power to be exercised 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. that supports the exercise of the Commission’s enforcement power explicitly referenced in the title of the respective article. As a consequence, it is reasonable to conclude that each of Articles 91(2), 92(7) and 93(2) AI Act confer 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. a power within the meaning of Article 53(3).

23If the discretion to initiate a structured dialogue is indeed a power, there is a duty on the part of the 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. to ‘cooperate as necessary’ in accordance with Article 53(3) AI Act. Intentionally or negligently⁵⁶ failing to satisfy the duty to cooperate would lead to an infringement of Article 53(3), which forms part of Article 53, titled ‘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’. Such infringement can be fined in accordance with Article 101(1)(a). This demonstrates the importance of considering what a 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. must do in practice to satisfy the duty to ‘cooperate as necessary’ in relation to structured dialogues.

24Once the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. initiates a structured dialogue, 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. is obliged to, at minimum, respond to the request. Given the lack of case law on Article 53 AI Act, its effect, and its interaction with the AI Act’s structured dialogue provisions at the time of writing (March 2026), it is instructive to consider how the CJEU has understood obligations to cooperate in other areas of EU law. In particular, in the area of competition law, the CJEU adopted a wide interpretation of an obligation to cooperate in Commission v SGL Carbon.⁵⁷ In this case, the CJEU considered the Commission’s exercise of a power to request information in relation to anti-competitive agreements pursuant to Article 11 of Council Regulation No 17. The CJEU held that there was ‘an obligation to cooperate actively’ which entailed an obligation to ‘make available all information relating to the subject-matter of the investigation’.⁵⁸ This decision can be distinguished from the structured dialogue scenario under the AI Act because it addressed the exercise of a key enforcement power: a request for information by decision. However, the tenor of the CJEU’s judgment suggests that, as a structured dialogue under the AI Act can only be a dialogue if 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. engages 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. , there is a good argument that the CJEU would require 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. to engage with and respond to a structured dialogue invitation in order to comply with its duty to cooperate.

25It is worth noting that such a requirement to engage and respond does not appear to offend the right to avoid self-incrimination recognised in EU law. In the context of decisions to request information, the case law of the CJEU suggests that undertakings cannot be forced to admit that they have committed an infringement, but they are, in any event, obliged to answer factual questions and to provide documents, even if this information may be used to establish against them or against another undertaking the existence of an infringement.⁵⁹ As in the previous paragraph, it is important to note that it remains unclear whether and how the case law on decisions to requests for information extends to structured dialogues. However, as requests for information do not allow an entity to evade requests for production of documents on the ground that, by complying with them, it would be required to give evidence against itself, an interpretation of structured dialogues that would require an entity to respond to an invitation for a structured dialogue is unlikely to offend against the right to avoid self-incrimination.

26This conclusion also seems relevant to understanding the operation of Article 93(3), which empowers 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. to offer commitments to implement mitigation measures during structured dialogues. Significantly, the Commission may make those commitments binding by decision and declare no further grounds for action. Drawing an analogy with the competition law context,⁶⁰ such commitments neither constitute an admission of guilt nor a formal finding of a violation under the AI Act.⁶¹ This mechanism offers what can be seen as a consensual enforcement regime, especially as the Commission may cease its enforcement actions if the commitments proposed by the respective 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. are accepted. This role of structured dialogues would be materially restricted if, while invoking the right to avoid self-incrimination, 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. were allowed to hold back information relevant to an infringement of the AI Act during structured dialogues.

27The duty to cooperate under Article 53(3) likely also entails a good faith obligation for GPAI 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. not to provide false or misleading information or intentionally delay or prolong a structured dialogue in order to delay regulatory investigations or actions. Firstly, such actions would be counter to teleological and literal interpretations of the words ‘cooperate as necessary’ under Article 53(3) AI Act. Secondly, Article 101(1)(b) enables the Commission to issue fines if a 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. supplied incorrect, incomplete or misleading information. This latter argument will be explored in the following subsection, as the good faith obligation not to provide false or misleading information arguably applies even if structured dialogues are found to not constitute a power and subsequently the duty of Article 53(3) is found not to apply to require a 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. to respond to, or engage 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. ’s initiation of a structured dialogue.

28It can be noted that recognition of a duty to cooperate at the stage of structured dialogues, in accordance with the analysis above, would distinguish structured dialogues from informal requests that have no legal effect, as well as from formal enforcement actions such as requests for documentation and information under Article 91(1).

29Finally, for the sake of completeness, we note that it could be argued that the power to initiate structured dialogues is separate from, and not per se part of, the ‘exclusive powers’ referred to in Article 88(1) AI Act, such that the implementation of such power could not, in principle, be conferred 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. . Article 91(2) AI Act provides that ‘[b]efore sending the request for information, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 initiate a structured dialogue with 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. […]’ (emphasis added). Similarly, Article 92(7) AI Act states that ‘[p]rior to requesting access to 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, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 initiate a structured dialogue with 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. […]’ (emphasis added). Likewise, Article 93(2) AI Act specifies that ‘[b]efore a measure is requested, the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 initiate a structured dialogue with 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. […]’ (emphasis added). In all three provisions, the structured dialogue is presented as an action that occurs prior to the exercise of the ‘power’ that is the central focus of the provision in question. A literal interpretation of these provisions might suggest that the structured dialogue is outside of the ‘power’ referred to in the title of each of these Articles. However, even if this is correct, the discretion to initiate a structured dialogue in each of the provisions conferring central investigative or enforcement powers on the Commission seems to fall within the implementation of the tasks conferred 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. to support the Commission’s ‘exclusive powers to supervise and enforce’ under Article 88(1). As elaborated earlier in this subsection, the discretion to initiate structured dialogues appears to be a supplementary power which ensures that the main enforcement power, named in the provision’s title, can reach optimal enforcement outcomes in various circumstances.

3.2. Interpretation 2: Structured dialogues are not a power for the purpose of Article 53(3) AI Act

30In contrast to the above subsection,⁶² there are also reasonable arguments that structured dialogues do not constitute a power. Three interrelated arguments are presented below. If these arguments present the correct interpretation, then the duty to cooperate under Article 53(3) AI Act arguably does not apply to structured dialogues. One potential implication may be that there is no obligation to enter into or respond to an invitation to a structured dialogue. The latter part of this subsection examines what duties may nevertheless still apply in relation to structured dialogues, in particular (as noted above) a good faith obligation that may arise under a broad interpretation of Article 101(1)(b) AI Act.

31One argument supporting the interpretation that the initiation of structured dialogues does not amount to a power arises from the fact that Article 101 AI Act does not expressly provide for fines in relation to engagement in a structured dialogue. Article 101 AI Act refers explicitly to Articles 91, 92 and 93 AI Act in three separate respective provisions. Fines may be imposed 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. that intentionally or negligently ‘failed to comply with a request for a document or for information pursuant to Article 91[…]’,⁶³ ‘failed to comply with a measure requested under Article 93’,⁶⁴ and ‘failed to make available to the Commission access to the [GPAI] model or [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. with a view to conducting an evaluation pursuant to Article 92’.⁶⁵ None of the provisions mentions structured dialogues. Thus, Article 101 AI Act does not give any suggestion that structured dialogues constitute a power and may be read to show that they do not. Further, on one interpretation, structured dialogues do not constitute an implicit part of the powers referred to in the titles of Articles 91, 92 and 93 AI Act. As analysed in the previous subsection,⁶⁶ this argument posits that the power to initiate structured dialogues could be seen as separate from the ‘powers’ referred to in the titles of Articles 91 to 93 AI Act on the basis that structured dialogues are to be initiated ‘before’ or ‘prior to’ (emphasis added) the ‘power’ that is to be exercised under the respective Article. This interpretation that structured dialogues sit outside the Articles 91 to 93 powers, in combination with the lack of express reference in Article 101 to structured dialogues, may suggest that the legislature did not intend for the initiation of structured dialogues to be conceived of as a power.

32While the lack of an express reference to the word ‘power’ in any of Articles 91(2), 92(7) or 93(3) AI Act seems inconsequential for characterising those provisions because the term is not always used in provisions that establish uncontested powers, the lack of indications that structured dialogues are mandatory processes seems more significant. It is worth highlighting that there is no explicit suggestion in the AI Act that 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. ‘must’ or ‘shall’ react in a certain way in cases where the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided for in Commission Decision of 24 January 2024; references in this Regulation to the AI Office shall be construed as references to the Commission. initiates a structured dialogue. The absence of punitive consequences if 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. fail to engage in a structured dialogue in accordance with Article 101 AI Act, addressed in the preceding paragraph, is also pertinent in this context. Further, the ordinary meaning of ‘dialogue’ as relating to the principle of free will⁶⁷ goes against the notion that ‘structured dialogues’ are mandatory.

33Relatedly, a third argument posits that the concept of structured dialogues is too vague to be considered a power. As elaborated throughout this chapter, much remains unclear about structured dialogues, including their purpose and character,⁶⁸ as well as the legal effects 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. using its discretion to initiate a structured dialogue.⁶⁹ Within the AI Act, this contrasts with other enforcement or investigative provisions, such as Article 91 AI Act, which empowers the Commission to request 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 to ‘provide the documentation drawn up 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. in accordance with Articles 53 and 55, or any additional information that is necessary for the purpose of assessing 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 this Regulation’. Further, Article 91(4) AI Act specifies that a ‘request for information shall state the legal basis and the purpose of the request, specify what information is required, set a period within which the information is to be provided, and indicate the fines provided for in Article 101 for supplying incorrect, incomplete or misleading information’.⁷⁰ However, it could be argued that Article 93 AI Act also lacks some specificity, with Article 93(1)(c) stating that ‘[w]here necessary and appropriate, the Commission may request 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. to […] restrict 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. , withdraw or recall the model.’ Taking a broader perspective of EU digital legislation, the vagueness of the provisions on structured dialogues contrasts with the provision on powers for independent supervisory authorities under Article 58 GDPR, which provides lists of ‘investigative powers’ and ‘corrective powers’ that each supervisory authority ‘shall have’.⁷¹ For example, according to Article 58(1)(e) GDPR, supervisory authorities shall have the investigative power ‘to obtain, from the controller and the processor, access to all personal data Article 3(50) AI Act: ‘personal data’ means personal data as defined in Article 4, point (1), of Regulation (EU) 2016/679. and to all information necessary for the performance of its tasks’. Article 58(2)(a) GDPR provides the corrective power to ‘issue warnings to a controller or processor that intended processing operations are likely to infringe provisions of this Regulation’. These are arguably more specific than ‘structured dialogues’.

34If a structured dialogue does not constitute a power, the duty to cooperate laid down in Article 53(3) arguably does not apply 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. uses its discretion to initiate a structured dialogue. If the duty to cooperate does not apply at that point in time, then there may be no obligation to enter into or respond to an invitation to a structured dialogue. If this is correct, there is conceivably an element of symmetry achieved: if the AI Office Article 3(47) AI Act: ‘AI Office’ means the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance, provided 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 obliged to initiate structured dialogues prior to adopting formal requests,⁷²  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. are not obliged to accept an invitation 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 enter a structured dialogue if received.⁷³

35If, however, 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. does choose to enter a structured dialogue, certain duties may arise. For the following reasons, it is suggested that one such duty could operate to prevent 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. from acting in bad faith during structured dialogues, for example, by providing false or misleading information.⁷⁴ A teleological interpretation of Article 101 AI Act could indicate a duty not to supply incorrect, incomplete or misleading information during the course of a structured dialogue. Article 101(1)(b) AI Act provides the possibility for the Commission to issue a fine if 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. intentionally or negligently ‘failed to comply with a request for a document or for information pursuant to Article 91, or supplied incorrect, incomplete or misleading information’ (emphasis added). Adopting a broad reading of Article 101(1)(b), which cannot be excluded at this stage in the absence of relevant case law, there is a reasonable argument that in this sentence the word ‘or’ is used as a disjunction, with the consequence that providing incorrect, incomplete or misleading information is a free-standing, separate action from not complying with a request under Article 91 and is susceptible to a fine under Article 101(1)(b).⁷⁵ So far, the Commission has not provided its own interpretation on this question.⁷⁶ It should be noted that this interpretation does not contradict the principle of free will as discussed in the literature:⁷⁷ 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. is free to decide whether or not to participate; however, if they do decide to participate, they must act in good faith.

36Lastly, it can be noted that regardless of whether structured dialogues constitute a power, there are pragmatic reasons that may de facto compel 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. to respond to structured dialogues. If a 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. does not engage in a structured dialogue, they may instead face the Commission exercising its powers under Articles 91 to 93. If anything, engaging in the structured dialogue process will either obviate the need for or allow a 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. to shape the Commission’s binding request.

37In light of the vagueness of the concept of structured dialogues as laid out in the AI Act, it is relevant to consider whether the Commission may further detail the concept in an implementing act under Article 101 AI Act.⁷⁸ Article 101(6) AI Act provides that ‘[t]he Commission shall adopt implementing acts containing detailed arrangements and procedural safeguards for proceedings in view of the possible adoption of decisions […]’.⁷⁹ Whether the empowerment under this provision requires the Commission to provide detailed arrangements for structured dialogues is ambiguous. The word ‘proceedings’ can be understood in several ways. On one reading, ‘proceedings’ are a unitary concept that encompass all aspects of an investigation using the powers laid down under Articles 91 to 93 AI Act. Under this reading, Article 101(6) AI Act would clearly empower the Commission to provide for ‘detailed arrangements and procedural safeguards’ for structured dialogues in an implementing act under Article 101(6) AI Act.

38On another reading, it may be argued that the AI Act contains (at least) two types of ‘proceedings’ in relation to enforcement of the obligations related to GPAI models, namely (i) investigative proceedings under Articles 91 to 93 AI Act and (ii) fining (or sanctioning) proceedings under Article 101 AI Act, with Article 101(6) AI Act only applying to fines. Arguably, this latter interpretation does not offer the Commission scope to provide detailed arrangements as regards investigative actions under Articles 91 to 93 AI Act in an implementing act adopted pursuant to Article 101(6) AI Act, including in relation to structured dialogues referred to in those investigative provisions. However, even this second, narrower interpretation would not eliminate structured dialogues entirely from the scope of the implementing act under Article 101(6) AI Act. This is because the Commission is required to take ‘into account commitments made in accordance with Article 93(3)’ during structured dialogues⁸⁰ for the purposes of setting an appropriate fine, per Article 101(1) AI Act.⁸¹ It follows that it is not possible to draw a sharp distinction between the consequences of a structured dialogue – that is, the binding commitment – and the fining decision. Therefore, it is foreseeable that the ‘detailed arrangements and procedural safeguards’ addressed by an implementing act adopted under Article 101(6) would need to engage with the role of a binding commitment under Article 93(3) AI Act in a fining decision.

39It could also be argued that it would be appropriate for an implementing act adopted under Article 101(6) AI Act to address structured dialogues in relation to Article 101(1)(b) following the analysis in the previous section on the duty that arises from Article 101(1)(b) AI Act to not provide incorrect, false or misleading information during structured dialogues. In particular, given that Article 101(1)(b) AI Act establishes liability for supplying incorrect, false or misleading information,⁸² and that such information may be provided in the context of structured dialogues, there is an arguable basis for including those dialogues within the procedural framework governing fines. A systematic and/or teleological reading of the AI Act supports an argument that the implementing act adopted under Article 101(6) AI Act should, at a minimum, articulate how participation in structured dialogues, especially the provision of information and the offering of commitments, interacts with the Commission’s powers to impose fines, offering procedural fairness.

40The procedural safeguards referenced in Article 94 AI Act, such as the right to be heard, only apply to a ‘measure, decision or order’ and thus at first sight exclude structured dialogues.⁸³ The absence of any applicable procedural safeguard would indicate that a response to a structured dialogue was not intended to be mandatory.

41However, it is likely that procedural safeguards do apply. Article 94 AI Act provides that the procedural safeguards set out in Article 18 of Regulation (EU) 2019/1020 apply on a mutatis mutandis basis. Consequently, it is arguable that even if a structured dialogue does not amount to a ‘measure, decision or order’, some procedural safeguards still apply. In any event, the Right to Good Administration contained in Article 41 of the Charter of Fundamental Rights is applicable and affords similar, and arguably additional, protections to those in Article 94.⁸⁴

42If there is the threat of a fine for providing incorrect or misleading information, the initiation of structured dialogues 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. should respect certain formal requirements. This follows from the principle of legal certainty, which aims to ensure that situations and legal relationships governed by EU law remain foreseeable.⁸⁵ This is particularly the case where the application of EU rules may have adverse consequences for a private person or entity.⁸⁶

43There is a measure of uncertainty as to the extent of procedural safeguards. It is suggested that the Commission should, by way of guidelines or implementing act, in particular clearly state the purpose of structured dialogues to distinguish them from purely informal exchanges and what is required from 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. .⁸⁷ As considered above, there is an argument that structured dialogues could fall within the scope of the implementing act to be adopted under Article 101(6) AI Act.⁸⁸ Even if that transpires not to be the preferred interpretation, as noted above, 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. are afforded procedural rights under Article 94 AI Act, without prejudice to more specific rights that are contained elsewhere in the legislation.⁸⁹ At a minimum, such procedural safeguards include a right to be informed of the grounds on which, inter alia, a decision is taken and the opportunity to be heard, save in exceptional circumstances.