What does the European AI Office do?

It supervises and enforces the general-purpose AI model obligations, develops codes of practice, templates and guidance, coordinates with national market-surveillance authorities, and supports the AI Board and the scientific panel of independent experts.

Does the AI Office enforce high-risk AI rules too?

High-risk AI systems are supervised primarily by national market-surveillance authorities. The AI Office's exclusive enforcement powers concern general-purpose AI models, though it coordinates across the system.

Are AI Office codes of practice mandatory?

No, they are voluntary, but adherence is a recognised way to demonstrate compliance with the corresponding obligations. A provider that does not follow a code must demonstrate compliance by other adequate means.

EU AI Act glossary · Governance

European AI Office

The European AI Office, within the European Commission, oversees the implementation and enforcement of the EU AI Act — especially for general-purpose AI models — and helps develop codes of practice and guidance.

The AI Office holds exclusive powers to supervise and enforce the GPAI obligations in Articles 53 to 55, including requesting documentation from model providers, conducting model evaluations, requiring mitigation measures and, where necessary, restricting a model's availability on the market. It also facilitates the drawing up of codes of practice, develops the templates the Act relies on — including the training-data summary template — coordinates with national market-surveillance authorities and supports the AI Board and the scientific panel of independent experts.

Where European AI Office sits in the Regulation

Art. 53
Art. 55
Art. 88
Art. 89

Who it binds

European AI Office: why it matters

The AI Office is the reason the GPAI layer behaves differently from the rest of the Act: model providers deal with one central regulator in Brussels rather than 27 national authorities, and that regulator both writes the guidance and enforces it. It is also the venue where much of the Act's practical meaning is being settled — codes of practice and templates fill gaps the legislative text left open, so a provider that ignores the Office's output is planning against an incomplete rulebook.

Worked example

European AI Office in practice

A GPAI provider preparing for Article 53 tracks the AI Office's template for the training-content summary and the code of practice on GPAI, because those documents determine the format its disclosures must take. If the Office later requests documentation or asks it to evaluate a capability, the request comes from Brussels directly — there is no national regulator in between, and no member-state forum in which to relitigate it.

Correction

European AI Office: the mistake people make

Assuming your national market-surveillance authority is the point of contact for everything. For high-risk AI systems it generally is, but for general-purpose AI models supervision sits exclusively with the AI Office, and mixing the two leads to notifications being sent to a body with no competence to receive them.

Questions

European AI Office: common questions

What does the European AI Office do?: It supervises and enforces the general-purpose AI model obligations, develops codes of practice, templates and guidance, coordinates with national market-surveillance authorities, and supports the AI Board and the scientific panel of independent experts.
Does the AI Office enforce high-risk AI rules too?: High-risk AI systems are supervised primarily by national market-surveillance authorities. The AI Office's exclusive enforcement powers concern general-purpose AI models, though it coordinates across the system.
Are AI Office codes of practice mandatory?: No, they are voluntary, but adherence is a recognised way to demonstrate compliance with the corresponding obligations. A provider that does not follow a code must demonstrate compliance by other adequate means.

Keep reading

European AI Office: terms that come up alongside it

GPAI model providerArt. 53–55

The model-layer role supervised by the AI Office rather than national authorities.

AI regulatory sandboxArt. 57

The supervised testing environment member states must offer, with SME priority.

Systemic risk (GPAI)Art. 51, 55

The compute-threshold tier that adds evaluation, incident and security duties to a GPAI model.

General-purpose AI (GPAI) modelArt. 3(63)

The foundation-model category with its own obligation set under Articles 53 to 55.

See how European AI Office applies to your systems

Run the free 2-minute EU AI Act checker — no sign-up.

Run the free check