The AI Office is established within the Commission and is the Union-level body for the supervision and enforcement of the AI Act in relation to general-purpose AI models and, following Regulation (EU) 2026/1744, specified AI systems assigned to Union-level supervision. It can investigate, request information and documentation, conduct evaluations, require corrective action and, under Article 101, impose fines on general-purpose AI model providers. It also supports the consistent application of the Act through guidance, coordination with national authorities and the Board, and work on codes of practice. It does not supervise every AI system: most systems remain under national market surveillance authorities.
Where does the AI Office sit, and what is it for?
The AI Office is established by Article 64 within the Commission, as a body that supports the development and coordination of AI policy and that carries out the supervision and enforcement tasks the Act assigns to the Commission. Its establishment reflects a design choice in the Act: general-purpose AI models are developed and supplied across borders, often by a small number of providers, so supervision at Union level avoids duplicating the same evaluation in every Member State. The AI Office also has a wider supporting role, contributing to guidance, standards work, codes of practice and coordination with the Board, the Scientific Panel and the Advisory Forum. That dual character, supervisor and coordinator, means its publications can matter in two ways: as enforcement action against a specific provider, and as implementation material that other operators and authorities rely on. Both kinds of publication should be read with their status and date attached.
What does the AI Office supervise?
The AI Office supervises providers of general-purpose AI models under Chapter V, including the obligations in Articles 53 and 55, the classification of models with systemic risk under Article 51, and the codes of practice under Article 56. Following Regulation (EU) 2026/1744, its competence was extended to specified AI systems, including general-purpose AI based systems and AI systems deployed on very large online platforms and search engines, with exclusive competence in defined respects that should be checked in the consolidated text. That extension is provision-specific and should not be read as a general transfer of all AI supervision to the Commission: high-risk systems, prohibited practices and transparency duties for most systems remain primarily with national authorities. A provider should therefore identify, for each model and system, whether supervision sits with the AI Office, a national authority, or both in different respects, and should record that allocation with the system's compliance file.
For the broader regulatory framework, see enforcement and market surveillance.
What powers can the AI Office exercise?
The AI Office can request information and documentation from general-purpose AI model providers, conduct evaluations of models, including evaluations of compliance with the Act's requirements, and require providers to take measures to address non-compliance. It can request access to the model, including through application programming interfaces or other technical means, and can require the provider to provide information about the model's development, testing and deployment. Where a provider fails to comply, the Commission, acting on the AI Office's work, can impose fines under Article 101 and periodic penalty payments under Article 101a, subject to the procedures and safeguards in the Act. Those powers are exercised under procedural guarantees, including the right to be heard and the protection of confidential information, and they are subject to judicial review. A provider facing an AI Office request should treat it as a formal supervisory step, with a named owner, a deadline diary and a complete response record.
How does the AI Office work with the Scientific Panel?
The AI Office works with the Scientific Panel, a body of independent experts established to support the implementation and enforcement of the Act, particularly in relation to general-purpose AI models with systemic risk. The Panel can provide advice, support evaluations and, through qualified alerts, draw the Commission's attention to a model that may present systemic risk. That alert mechanism matters because it gives an expert route into supervision that does not depend on the provider's own notification. The AI Office can also draw on the Panel's expertise when assessing a model's capabilities, risks and mitigations, and the Panel's advice should be distinguished from a legal determination: it informs the Commission's decision, but the decision and its reasoning remain the Commission's. A provider should therefore expect that its model may be assessed with independent expert input, and should keep evaluation evidence in a form that can be shared and tested.
What guidance and codes does the AI Office produce?
The AI Office supports the development of codes of practice for general-purpose AI models, guidance on the Act's implementation and coordination of the consistent application of the Act across Member States. Codes of practice are voluntary instruments that can help a provider demonstrate compliance with Chapter V obligations, and adherence to an applicable code can be taken into account in enforcement, but a code is not itself the law and does not replace the obligations in the Regulation. Guidance documents should be cited with their date and status, and draft guidance should be labelled as draft. A provider should track AI Office publications that concern its models, record which publications were relied on for which decision, and update its position when a final code or guideline replaces a draft. Relying on a superseded draft, or treating a voluntary code as a safe harbour, are both errors the record should be designed to prevent.
Continue with fines for general-purpose AI providers under Article 101.
What does this guide not decide?
This guide does not decide whether a particular model or system falls under AI Office supervision, what the AI Office will require in a particular evaluation, or whether a provider has complied with Chapter V. Those questions depend on the model, the provision and the consolidated text as amended by Regulation (EU) 2026/1744, and they require the specific facts to be examined. The guide also does not describe the AI Office's internal organisation or staffing, which are operational matters for the Commission. Where AI Office supervision, an information request or an evaluation is in issue, confirm the position with the AI Office or a legal adviser, and keep the request, the response and the evidence with the model's compliance record.
Frequently asked questions
Does the AI Office supervise all AI systems?
No. It supervises general-purpose AI models and specified systems assigned to Union-level supervision. Most AI systems, including most high-risk systems, are supervised by national competent and market surveillance authorities.
Can the AI Office impose fines?
Fines for general-purpose AI model providers under Article 101 are imposed by the Commission, acting on the AI Office's supervisory work, subject to the procedures and safeguards in the Act.
Does following a code of practice guarantee compliance?
No. Codes of practice are voluntary instruments that can support a demonstration of compliance and can be taken into account in enforcement, but they do not replace the obligations in the Regulation itself.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689, as amended ↗AI Act Service Desk implementation timeline ↗