EU AI Act GPAI Code of Practice published in final form — May 28 release references three-lab procedural convergence, enters force August 2026
The European Commission published the final form of the EU AI Act General-Purpose AI Code of Practice on May 28. The document references the three-lab procedural convergence on capability-driven release gating (Anthropic Mythos, OpenAI Superalignment Report, DeepMind Frontier Safety Framework v3) as the operational baseline for the systemic-risk evaluation requirements, and enters force in August 2026. The publication is the most consequential single AI-policy artifact of the cycle.
The Code of Practice content is the substantive piece. The document — running approximately 220 pages with annexes — covers four major sections: the model-evaluation requirements for systemic-risk GPAI providers, the documentation-and-transparency requirements for downstream deployers, the incident-reporting and post-market monitoring obligations, and the governance procedure linking provider-level evaluations to Commission-level oversight. The model-evaluation section explicitly references the methodological frameworks Anthropic, OpenAI, and DeepMind have published — meaning the regulatory requirements piggyback on the procedural convergence the frontier labs have already produced rather than specifying capability-evaluation methodology from scratch.
The timing and enforcement context is what makes the publication consequential. DeepMind's FSF v3 release the same day completes the three-lab procedural alignment the Commission relies on; OpenAI's Superalignment Report is the other pillar. The Code of Practice enters force in August 2026 with a 12-month transition period for systemic-risk GPAI providers to come into procedural compliance. For the major frontier labs, the procedural compliance work is substantially already done — the published methodologies map onto the Commission requirements with relatively minor gaps. For smaller GPAI providers and the broader open-source ecosystem, the compliance work is more substantial; the document includes proportionality provisions that scale obligations to deployment context and capability tier.
European Commission — AI Act GPAI Code of Practice final publication May 28 2026 → · Euractiv — EU AI Act Code of Practice references frontier-lab frameworks → · Reuters — EU GPAI Code of Practice enters force August 2026 →