The AI Office: role, tasks and enforcement powers

Short answer: The AI Office is a department within the European Commission that centrally coordinates implementation of the AI Act. It has one unique, exclusive task: direct supervision and enforcement of rules for general-purpose AI (GPAI) models. For "ordinary" AI systems the powers lie with national supervisors; the AI Office safeguards consistency.

What the AI Office is

The AI Office is not a separate agency but part of the Commission (DG CNECT). It was set up to make the AI Act land uniformly across the Union: pooling technical expertise, supporting Member States and acting where European coordination is needed. It works with the Member States' AI Board, a scientific panel of independent experts and a broader advisory forum.

Exclusive supervision of GPAI models

The AI Office's sharpest power concerns the GPAI regime: large models such as the generative language models behind many applications. Here the AI Office is not one of several supervisors but the only one. It can:

• request documentation and information from model providers;
• carry out (or commission) model evaluations, including to assess systemic risk;
• require measures, impose restrictions or have a model withdrawn from the market;
• advise the Commission on fines of up to 3% of worldwide annual turnover or EUR 15 million for GPAI providers.

Centralising model supervision prevents 27 supervisors from separately assessing the same few large providers.

Coordination, codes of practice and guidelines

Beyond enforcement, the AI Office has a standard-setting role. It organises the drafting of codes of practice that GPAI providers can use to show they meet their obligations, and it prepares guidelines and delegated acts that fill in the open norms of the AI Act. It also supports alignment among national supervisors so the same rule is not interpreted differently in each Member State.

Limits of its remit

The AI Office does not enforce everything. For high-risk AI systems and prohibited practices, the national supervisors are competent; they handle complaints and impose fines on providers and deployers. The AI Office mainly guards the model layer and the unity of interpretation. To know where a specific complaint belongs, first determine whether it concerns a model or an application — see filing a complaint.

What to do

• Identify your role in the chain: are you a GPAI provider (AI Office) or a deployer of an application (national supervision)?
• Follow the codes of practice: for GPAI providers these are the practical compliance benchmark.
• Track the guidelines: the AI Office fills in open norms over time; those documents shape practice.
• Keep model documentation: the AI Office can request it directly.
• Read about fines and enforcement to understand the financial stakes.