a daily news desk
Deployments

EU AI Office switches on enforcement, and every chatbot in Europe now has to say it's a bot

Article 50 transparency duties and GPAI fines up to €15M or 3% of turnover took effect Sunday, with OpenAI, Anthropic and Google already in scope.

Two years to the day after the AI Act entered into force, the European AI Office switched its enforcement powers on. As of 2 August 2026, the Commission can inspect general-purpose AI models before release, restrict market access, and fine providers up to €15 million or 3% of worldwide annual turnover, whichever is higher. Article 50’s transparency regime went live the same day: any system talking to a user has to say it’s a machine, and generative output has to carry machine-readable provenance marks. Content already shipped gets a grace period until 2 December to be labelled.

The Commission isn’t warming up on hypotheticals. CNBC reports it’s already in talks with OpenAI and Anthropic over recent cyber incidents involving their models, with Google inside the same perimeter. Elisabetta Righini of Sidley Austin notes that “a U.S. address does not put a lab outside the EU regulator’s reach,” and that GPAI liability “isn’t limited to substantive breaches: refusing an information request, giving misleading answers, or blocking a model evaluation is fineable on its own.” Process becomes substance.

The perimeter is the real story. Cooley reads the territorial scope as covering any provider or deployer whose AI outputs are consumed in the EU, which sweeps in every SMB running a customer-facing agent through a no-code platform like Glean, Dust, or LemonLime. Wilson Sonsini expects technical compliance dialogues to “intensify” before formal powers get used in anger.

Not everyone is comfortable with how the sequencing looks. The Center for Democracy & Technology, quoted by Euronews, warned the Commission not to let enforcement go “headline-driven” toward cyber-offence cases at the expense of fundamental-rights risk. It’s a familiar critique of European tech regulation: the first cases set the doctrine, and the first cases are usually the ones the press is already watching.

Sources