European AI Office

The European Commission's centre for AI. It enforces the AI Act's rules for the most powerful general-purpose AI models, including testing them for risks such as loss of control.

The AI Office sits within the European Commission and helps implement the AI Act. It “enforces the rules for GPAI models” (general-purpose AI): it can request information from their makers, evaluate the models’ capabilities, and require fixes, restrict a model or fine its maker. It also supports the AI Board of national authorities and works on international agreements. European Commission

It drew up the General-Purpose AI Code of Practice (2025) with AI developers and experts. Its safety chapter lists four systemic risks for the most capable models, including “loss of control”: humans “losing the ability to reliably direct, modify, or shut down a model”, for example through deception, self-improvement or resistance to goal modification. Appendix 1.4