EU delays high‑risk AI rules

The European Parliament and the Council of the European Union reached a provisional political agreement on May 7 to revise the bloc’s AI Act, pushing back the toughest compliance dates for high-risk systems while keeping broad transparency duties in place. The Council said the deal would move the deadline for stand-alone high-risk AI systems to December 2, 2027, and for high-risk AI embedded in regulated products to August 2, 2028. The amendments are part of a wider simplification push in Brussels and still need formal approval by both institutions. The underlying AI Act, adopted in 2024, remains in force. ### Which parts of the AI Act were actually delayed? The Council said the delay applies to the rules for high-risk AI systems, not to the entire regulation. Under the provisional deal, stand-alone high-risk systems listed under Annex III would move from the original August 2026 timetable to December 2, 2027, while high-risk AI that is a safety component of regulated products would apply from August 2, 2028. (consilium.europa.eu) Regulation (EU) 2024/1689 set out a staggered rollout from the start, with different obligations applying on different dates. The May 7 deal changes those dates for the high-risk categories but does not replace the broader structure of the law. ### What did Brussels leave untouched on transparency? Article 50 of the AI Act still reaches beyond the high-risk category. (consilium.europa.eu) The text requires providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text content to ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. (eur-lex.europa.eu) The same article also requires certain deployers to inform people when they are exposed to emotion-recognition or biometric categorization systems, and to disclose when users are interacting with an AI system unless an exception applies. Those duties are framed as transparency obligations, not as part of the high-risk regime alone. (artificialintelligenceact.eu) ### Why are lawyers and companies focusing on Article 50? Article 50 matters because it captures systems that may never be classified as high-risk. A company offering chatbots, image generators or tools that manipulate media may still face labeling and detectability requirements even if the later high-risk deadlines do not apply to its product. (eur-lex.europa.eu) The Council also said the provisional agreement reinstates an obligation for providers to register some systems in the EU database for high-risk systems when they consider those systems exempt from high-risk classification. That keeps an administrative reporting element in the framework even as the main compliance timetable shifts. (artificialintelligenceact.eu) ### What does the provisional deal change for companies right now? May 7 did not erase current obligations. Companies still need to track which parts of the AI Act already apply, which systems could fall under Article 50, and which products may later be treated as high-risk under Annex III or under sectoral product rules. (consilium.europa.eu) The most immediate practical question is inventory. Providers and deployers need to identify where AI systems interact with users, where synthetic content is generated, and whether outputs can be marked and detected in a machine-readable way, because those transparency duties are written into the law’s operative text. (consilium.europa.eu) ### What still has to happen before the new dates are final? Formal adoption by the European Parliament and the Council is the next step. The Council described the May 7 outcome as a provisional agreement, which in EU legislative practice means negotiators have settled the text politically but the institutions still need to approve it through their formal procedures. (artificialintelligenceact.eu) The next milestones are the Parliament vote, the Council’s final sign-off and publication of the amending text in the Official Journal. Until those steps are completed, companies have a political agreement to plan around, but the legal text to watch remains the AI Act and the final amending measure once adopted. (consilium.europa.eu)