EU AI Act Omnibus political agreement May 7 extends high-risk deadlines — formal adoption expected July ahead of August 2 GPAI window
EU legislative bodies reached political agreement on May 7, 2026 on proposed amendments to the AI Act (the "AI Omnibus") that extend compliance deadlines for high-risk AI systems and introduce new rules on AI-generated intimate content. Formal adoption by the European Parliament and Council is expected by July 2026, ahead of August 2 when high-risk AI systems requirements would otherwise take effect.
The substantive piece is the bifurcation between high-risk delay and GPAI proceed. The Omnibus delays high-risk classification deadlines but does NOT delay the August 2 GPAI window — GPAI providers must still publish training-data summaries, copyright-compliance documentation, and model cards on the original schedule. For OpenAI, Anthropic, Google, and Meta, the practical compliance posture is unchanged: ship GPAI disclosures August 2, prepare for high-risk classification work on the extended timeline.
The political-economy frame is that the delay primarily benefits the industries that integrate AI into regulated products (medical devices, automotive, infrastructure) — not the frontier labs themselves. The June 10 Code of Practice on AI-generated content marking sits on the parallel track: GPAI is the model-supply side; content marking is the deployer-supply side; both lock in August.
Latham & Watkins — AI Act Update: EU Resolves to Change Rules and Extend Deadlines → · European Commission — AI Act | Shaping Europe's digital future → · IAPP — EU AI Act reform talks stall as key compliance deadline looms →