- The administrative constellation of the EU AI Act
The EU Artificial Intelligence Act, adopted on 13 June 2024, is a landmark policy for the regulation of AI. In essence, the AI Act adopts a risk-based approach, introducing escalating obligations depending on the risks associated with general purpose AI models (GPAIMs) or the relevant AI system. Substantive provisions are coupled with an elaborate structure of administrative enforcement, both on the EU and national levels.
Before setting out this structure, the scope of application of the AI Act must be acknowledged. According to Article 2 thereof, the Act applies to, inter alia, providers of AI systems or GPAIMs ‘placing’ them ‘on the market […] in the Union, irrespective of whether those providers are established or located within the Union or in a third country’. Given that extraterritorial effect, the enforcement mechanism envisaged under the Act has a real global salience, going beyond its strictly European relevance.
The AI Act provides for three main stakeholders in the enforcement and governance of the applicable rules: first, the AI Office; second, the European AI Board; third, national competent authorities in member states. The AI Act requires and relies on their cooperation. The following section sets out the tasks of each of the three enforcement stakeholders and identifies ways in which they must cooperate inter se.
First, as suggested by Article 64 AI Act, the AI Office was established within the European Commission. The AI Office is not a distinct entity but, rather, ‘the Commission’s function of contributing to the implementation, monitoring and supervision of AI systems and general-purpose AI models, and AI governance’ (Article 3(47) AI Act). Specifically, it forms part of the Directorate-General for Communication Networks, Content and Technology.
The tasks of the AI Office are more fully set out in the Commission Decision providing for its establishment. Inter alia, it will monitor the implementation and application of applicable rules on GPAIMs and AI systems; monitor the emergence of ‘unforeseen risks’; and investigate ‘possible infringements’ of applicable rules (Article 3(1) Decision). The AI Office has a particularly pronounced role in relation to GPAIMs: Article 88 AI Act asserts its ‘exclusive powers to supervise and enforce’ rules concerning GPAIMs, whereas Article 92 AI Act sets out the Office’s power to conduct ‘evaluations’ of GPAIMs, after consulting the Board, with a view to assessing providers’ compliance with applicable obligations and investigating ‘systemic risks’ ex ante. Under Article 93, the Office can further request measures to ensure compliance or risk mitigation.
Importantly, the AI Office will ensure that the Commission’s supervision and enforcement under the Digital Markets Act and the Digital Services Act is ‘fully coordinated’ with that under the AI Act. More broadly, the AI Office has a policy-making function, as it must, for instance, ‘contribute to the strategic, coherent and effective Union approach to international initiatives on AI […] in coordination with Member States and in line with Union positions and policies’ (Article 2(2)(a) Decision). Finally, the Commission has recognised the importance of cooperation with various national, EU and international stakeholders as essential in conducting the tasks of the AI Office (Articles 4-7 Decision).
Under Articles 67 and 68 AI Act, the Commission shall further establish an Advisory Forum and a Scientific Panel. These bodies are envisaged to provide expert input in the governance and enforcement tasks of the Commission and, where appropriate, the AI Board.
Second, Article 65 AI Act established the European AI Board. The Board is composed of one representative by each member states. The European Data Protection Supervisor participates as an observer, and the AI Office shall attend (without the right to vote) the Board’s meetings and provide administrative support. The Board is overall responsible for ‘advis[ing] and assist[ing]’ the Commission and the member states ‘to facilitate the consistent and effective application’ of the AI Act (Article 66 AI Act).
This entails a wide range of more specific tasks. The Board shall, firstly, draw from its EU-wide composition: for instance, it will contribute to the coordination of national competent authorities, serve as a forum for sharing expertise and best practices, and ‘contribute to the harmonisation of administrative practices’ across the member states. Secondly, the Board has an advisory role: for instance, it will ‘provide advice’ on the implementation of AI Act, particularly in relation to GPAIMs, will ‘issue recommendations and written opinions on any relevant matters’ pertaining to the implementation and application of the AI Act, contribute to the development of ‘guidance documents’, and advise the Commission concerning relevant ‘international matters’. Thirdly, the Board has a knowledge dissemination and capacity-building function: for instance, it will ‘support’ the Commission inter alia in ‘promoting AI literacy’ on the implications of using AI systems, and ‘assist the AI Office’ in ‘supporting’ national competent authorities in developing AI regulatory sandboxes and ‘facilitat[ing] cooperation and information-sharing’ between them.
Third, the role of national competent authorities (NCAs) in each member state is crucial for the enforcement of the AI Act. Under Article 70(1), each member state must designate at least one NCA as notifying authority and at least one as market surveillance authority. Notifying authorities are ‘responsible for setting up and carrying out the necessary procedures for the assessment, designation and notification of conformity assessment bodies and for their monitoring’ (Article 28 AI Act). Bodies which satisfy requirements under Articles 30 and 31 may, as notified bodies, assess the compliance of high-risk AI systems with obligations under the Act.
Market surveillance authorities act, within their respective member state, as the designated single point of contact for the AI Act (Article 70(2) AI Act). These authorities are responsible for surveilling national markets, in line with previous EU legislation, particularly in relation to high-risk AI systems (Article 74 AI Act). It is recalled that the Board will have a strong coordinating role in this regard. Subject to ‘exceptional reasons’ of certain public interests, such as ‘public security’, they are able to derogate from the conformity assessment procedure and authorise the placing on the marker of some high-risk AI systems within the relevant member state (Article 46 AI Act). They shall also maintain a close relationship with national public authorities which ‘supervise or enforce’ fundamental rights under EU law, including non-discrimination, in relation to high-risk AI systems— in this context, competent authorities can request access to any relevant documentation from the market surveillance authority (Article 77 AI Act).
- Drawing lessons from competition law cooperation
As AI systems advance at an unprecedented pace, cooperation among these authorities becomes essential for the effective implementation of the AI Act. In this context, it is beneficial to draw lessons from existing cooperative frameworks in other areas of law. The field of competition law offers valuable insights for fostering administrative cooperation among AI authorities. The experiences of competition authorities can help identify effective strategies to address similar challenges that AI authorities may encounter during the implementation phase.
One of the most pressing challenges include the risk of under-enforcement and inconsistent outcomes in similar or related cases across different jurisdictions. A primary source of these challenges is the lack of information sharing. In extreme cases, one authority may be unaware of crucial information held by another authority in a different jurisdiction handling the same or related case. This scenario underscores the importance of administrative cooperation and highlights a key topic: effective collaboration relies on the ability to exchange information. Therefore, any form of cooperation must include mechanisms to facilitate and protect this exchange.
Both formal and informal mechanisms coexist in competition law to address these challenges, and the key lies in identifying the most effective ways to use them. Competition law uses formal mechanisms, such as binding regulations, to enable the exchange of confidential information, and informal mechanisms, like non-binding forums, to provide platforms for sharing experiences and shaping policy. For instance, Regulation 1/2003 established specific rules and safeguards for information exchange, including the creation of the European Competition Network (ECN), a network of Member States’ competition authorities designed to facilitate discussion and cooperation in the application and enforcement of competition policy within the EU. An example of an informal mechanism is the International Competition Network (ICN), which offers non-binding recommendations, toolkits, and reports to promote the exchange of best practices.
Striking an appropriate balance between formal and informal mechanisms is essential for effective cooperation. This is evident in the ongoing discussions about the role of the ECN and the potential for reforms toward binding rules. The ECN issues non-binding recommendations, which are characterized as soft law. However, these recommendations do not provide legal certainty and cannot replace clear, binding legal standards that ensure predictability for stakeholders. As territorial boundaries become increasingly blurred, the demand for binding legal standards intensifies, underscoring the urgent need to reevaluate the role of the ECN in this context.
AI authorities will, in many ways, follow a path similar to that of competition authorities and should leverage their experiences. The AI Act includes provisions on information exchange and confidentiality that should be complemented by more specific rules to guarantee formal exchanges, similar to those in Regulation 1/2003. The AI Act also established an AI Board akin to the ECN, designed to coordinate and ensure cooperation among EU Member States. This board’s role should be subject to ongoing evaluation and adaptation to address emerging challenges in the field of AI. Furthermore, mechanisms for cooperation should reflect insights from competition authorities regarding the principle of “the more the merrier,” or the network effect, which suggests that the effectiveness of cooperation increases with the number of entities involved. Overall, AI authorities should embrace both formal and informal cooperation, drawing from the experiences of competition law.
Ultimately, as AI continues to reshape our society, the importance of cooperation among authorities cannot be overstated. By prioritizing information sharing and adopting best practices from fields like competition law, AI authorities can navigate the complexities of regulation more effectively and pave the way for a balanced approach to innovation and accountability.
Bibliography
Błachucki, M. (2023). International Cooperation, Competition Authorities and Transnational Networks (1st ed.). Routledge. https://doi.org/10.4324/9781003376002
Błachucki, M. (2020). On the phenomenon of international cooperation: From networks of authorities to networks of people. In M. Błachucki (Ed.), International cooperation of competition authorities in Europe: From bilateral agreements to transgovernmental networks (pp. 19–24). Warsaw: Publishing House of ILS PAS. Available at SSRN: https://ssrn.com/abstract=3751630
Csatlós, E. (2018). The European Competition Network in the European Administrative System: Theoretical Concerns.
Kloza, D., & Mościbroda, A. (2014). Making the case for enhanced enforcement cooperation between data protection authorities: Insights from competition law. International Data Privacy Law, 4(2), 120–138. https://doi.org/10.1093/idpl/ipu010
Maher, I., & Hosseini, M. (2024). Cooperation between national competition authorities: An overview of EU and national case law. e-Competitions Cooperation between National Competition Authorities, Article N° 116089.