European Commission: AI Act, AI Office & Enforcement Role

The only body that can propose EU law

The European Commission is the executive arm of the European Union, made up of one commissioner per member state plus a president, currently Ursula von der Leyen, who has held the role since December 2019. Roughly 32,000 staff across more than 30 departments run day-to-day EU administration, but the Commission's distinct power is legislative initiative: it is the only EU institution that can propose new legislation for the European Parliament and the Council to adopt. Once a law is adopted, the Commission is also the body that enforces it, which is what makes it the relevant actor for AI Act obligations rather than the Parliament that voted the Act into law. [2]

The AI Office enforces the rules for models

The Commission created a dedicated AI Office to administer the AI Act, with more than 125 staff across six units covering areas from AI safety to international coordination. For general-purpose AI models, the kind ChatGPT, Claude and Gemini are built on, the AI Office is the enforcement body: it develops methods for evaluating a model's capabilities, classifies which models carry systemic risk, can investigate suspected violations, request corrective action, and impose fines directly rather than routing enforcement through national regulators. It also convenes the European Artificial Intelligence Board of member-state representatives and an advisory forum of industry, academic and civil-society voices. [3]

The Commission on the timeline

The two events credited to the European Commission mark the points where the AI Act's text turned into obligations model providers actually had to meet. In August 2025, the general-purpose AI rules took effect: technical documentation, a public summary of training data, and a copyright policy, with models already on the market given until August 2027 to comply. Models that came to market after that date had to comply immediately. [4]

A year later, the Commission began enforcing the Act's transparency rules in August 2026: chatbots have to disclose that they are AI, deepfakes have to be labelled, and AI-generated content has to carry a machine-readable mark. By this point the Act's harder high-risk obligations had already slipped to 2027 and 2028, so what actually changed on this date was that the AI Office held real enforcement power for the first time, not that a new category of rule appeared. [5]