European Commission details AI Act investigative powers and penalties

On October 6, 2026, the European Commission published an enforcement-framework overview confirming the AI Office’s authority to issue information requests, require model access and evaluations, inspect providers and, where necessary, restrict a model’s public availability. The page states that enforceable breaches can attract penalties of up to €35 million or 7% of worldwide annual turnover for prohibited practices, with separate maxima for other violations and deficient responses to information requests. Law firms advising providers or deployers should now treat model documentation, downstream-provider records and regulator-response protocols as litigation-ready evidence rather than compliance paperwork. Legal departments procuring general-purpose models should also map who can supply technical information when the AI Office or a national authority asks for it.