What was decided
On June 29 the Council formally adopted the package that Parliament and member states agreed on in May. The core change: obligations for high-risk AI systems under Annex III, originally due on August 2, 2026, now apply sixteen months later. The omnibus also simplifies documentation duties for smaller providers and streamlines the conformity assessment process.
Not everything moved
The delay is targeted, not general. Prohibited practices remain banned, and the rules for general-purpose AI models that took effect in 2025 stay in place. New prohibitions agreed in the package take effect in December 2026, so the direction of travel is unchanged: more time, not less regulation.
What it means for founders
If you are building or deploying AI that touches hiring, credit, education, medical devices or critical infrastructure, you just received breathing room, not a pardon. Use it. Companies that treat the new deadline the way most treated the old one, as a problem for later, will find that enterprise customers ask for AI Act readiness in procurement long before regulators do. Compliance is becoming a sales document before it is a legal one.