EU Activates AI Act Obligations for General Purpose AI Models

The European Union Artificial Intelligence Act reached its first major enforcement milestone as obligations on providers of General Purpose AI models — defined as models trained on more than 10 to the power of 25 floating point operations — formally took effect across all 27 member states. Providers including OpenAI, Google, Meta, and Anthropic must now maintain technical documentation, publish model capability summaries, implement copyright compliance measures, and undergo adversarial testing by approved third-party auditors. Models classified as presenting systemic risk — those exceeding 10 to the power of 26 FLOPs — face additional incident reporting requirements. Non-compliance can attract fines of up to 3% of global annual turnover. The AI Office within the European Commission has begun its first formal assessments of the largest frontier models, with initial findings expected in early 2026.