European Union member states and the European Parliament agreed early Thursday to push back the toughest deadlines under the bloc’s AI Act, giving enterprises more time to prepare for high-risk compliance.
Under the provisional deal between negotiators for the European Parliament and European Council, high-risk AI systems will face new deadlines of Dec. 2, 2027 for stand-alone systems and Aug. 2, 2028 for AI used in products covered by EU sectoral safety rules, a European Parliament statement said. The original deadline was Aug. 2, 2026.
The deal still needs formal adoption by both Parliament and Council before it can enter into law. The co-legislators intend to complete that step before Aug. 2. Until they do, the original deadline applies as drafted.
“Today’s agreement on the AI Act significantly supports our companies by reducing recurring administrative costs,” Marilena Raouna, Cyprus’s deputy minister for European affairs, said in a statement from the Council, which is composed of representatives of each of the EU’s 27 member states. Cyprus holds the rotating presidency of the Council, which negotiates on behalf of member states.
The breakthrough comes nine days after previous discussions collapsed without agreement.
Fewer restrictions, more time to implement
The provisional agreement removes overlapping rules for AI in machinery products, Parliament said. These will now follow only sectoral safety rules, with safeguards meant to ensure equivalent health and safety protection.
It also narrows what counts as a “safety component” under the AI Act. AI features that only assist users or improve performance will not automatically be treated as high-risk, the Parliament said, as long as a failure does not create health or safety risks.
For wider sectors such as medical devices, toys, lifts, machinery and watercraft, the co-legislators agreed on a mechanism to resolve overlaps between the AI Act and existing sectoral laws, the Council said in its statement.
The deadline for member states to set up AI regulatory sandboxes has been pushed back by a year to Aug. 2, 2027, the Council said. Watermarking obligations on AI-generated content, on the other hand, will apply earlier than the Commission proposed, from Dec. 2, 2026 instead of Feb. 2, 2027, the Parliament said.
Mid-size firms get more breathing room. Exemptions previously available only to small and medium-sized enterprises now extend to small mid-cap companies, the Council said. The deal also clarifies that the EU’s AI Office will supervise general-purpose AI systems centrally, with national authorities keeping responsibility in areas including law enforcement, border management, judicial authorities and financial institutions.
“With this agreement, we show that politics can move just as quickly as technology,” said Arba Kokalari, the Parliament’s co-rapporteur for the Internal Market and Consumer Protection committee. “We now make the AI rules more workable in practice, remove overlaps and pause the high-risk requirements.”
Parliament and Council also agreed to ban AI systems that create child sexual abuse material or that depict identifiable people in sexually explicit content without consent, the Parliament said. The ban covers placing such systems on the EU market, doing so without safety measures to prevent misuse, and using them to generate the content. Companies have until Dec. 2, 2026 to comply.
“Alongside simplification measures, we are banning nudification apps, a key part of the Parliament’s mandate, and, of course, the creation of child sexual abuse material using AI systems,” said Michael McNamara, the Parliament’s co-rapporteur for the Civil Liberties, Justice and Home Affairs committee.
What still applies
Several parts of the AI Act keep moving on their original schedule. Bans on unacceptable-risk AI have applied since February 2025, according to the European Commission. The general-purpose AI rules came into force in August 2025. The transparency obligations under Article 50, including disclosure for chatbot interactions, are set to apply from Aug. 2, 2026.
The provisional agreement is part of the seventh omnibus package on simplification, proposed by the Commission on Nov. 19 last year in response to the Draghi report on EU competitiveness.
