EU AI Act omnibus reaches political agreement — high-risk rules pushed to Dec 2027
The 'AI omnibus' (proposed November 2025) reached political agreement on May 7, 2026. The practical effect: rules for high-risk areas — biometrics, critical infrastructure, education, employment, migration, asylum and border control — now apply from December 2, 2027, rather than the originally scheduled 2026 dates.
The pushback was always going to happen — compliance infrastructure for the original timeline didn't exist on the ground at the Member State level. The omnibus formalizes that reality rather than maintaining a deadline nobody could meet. Prohibited practices (entered into force February 2025) and GPAI obligations (August 2025) remain on their original schedules.
For frontier model providers, the immediate effect is more time to implement systemic-risk evaluations and incident reporting under the GPAI rules, which the omnibus did not touch. For sectoral deployers (HR tech, biometric ID, etc.), it's 14 extra months. The wider European AI policy posture remains: precautionary on high-risk deployment, more permissive on foundation-model training itself.