The Council of the EU reached a political agreement on March 13, 2026, to push back the EU AI Act’s high-risk AI system compliance deadlines by 16 months. According to Compliance Week’s reporting on the Council agreement, two new deadlines have been proposed: December 2, 2027 for stand-alone high-risk AI systems, and August 2, 2028 for high-risk AI systems embedded in regulated products. Neither date is final.
That last point matters. As Tech Policy Press notes, the delay must be formally enacted before the original August 2, 2026 deadline to take legal effect. The March 13 agreement is a Council position, not a legislative amendment. A formal political agreement must be reached before June 2026 to clear the procedural timeline. If it doesn’t clear, the original deadlines remain in force.
The EU AI Act structures its high-risk category around two distinct tracks, and the proposed extension treats them differently for good reason. Stand-alone high-risk systems, AI used independently in areas like employment, education, and credit scoring, would move to December 2027. Embedded systems, AI built into medical devices, vehicles, and other regulated products already governed by EU product safety law, would get until August 2028. The longer runway for embedded systems reflects the additional complexity of dual-compliance: these providers must satisfy both the AI Act and the relevant product regulation simultaneously.
The same Council agreement reportedly added a new prohibition. According to Compliance Week’s reporting on the agreement, a provision banning AI generation of non-consensual sexual material and child sexual abuse material was included alongside the deadline changes. This claim couldn’t be independently corroborated in this verification cycle, but it signals the direction the Council is moving: more time on high-risk obligations, no flexibility on content safety.
Compliance Week also reported that on March 11, 2026, the European Parliament approved the Council of Europe’s Framework Convention on AI, described as the first legally binding international AI governance treaty. That claim also awaits independent corroboration, but if confirmed, it adds a second governance layer that EU-market organizations will need to track alongside the AI Act itself.
For compliance teams, the practical picture looks like this. The staggered implementation framework the EU AI Act already established, with prohibited systems rules already in force, GPAI model obligations following, and high-risk deadlines trailing, now has two proposed anchor points roughly 18 months further out than originally scheduled. The EU’s official digital strategy documentation confirms the staggered structure. The proposed dates give teams breathing room on classification, conformity assessment, and documentation requirements. What they don’t change: anything already in force, and any new obligations added alongside the extension.
The watch item is June 2026. That’s the practical deadline by which formal enactment needs to move if the proposed extensions are to replace the originals. Organizations already treating August 2026 as a live deadline should hold that posture until a formal agreement is published. The Council’s political agreement is a strong signal, not a guarantee.