EU delays AI Act obligations

Europe’s AI rulebook is still coming. But the part many companies feared most just got pushed back. On May 7, negotiators from the Council of the EU and the European Parliament struck a provisional deal to delay key obligations for high-risk AI systems, while also adding a new explicit ban on so-called “nudifier” apps. (consilium.europa.eu) ### What part of the AI Act got delayed? The delay hits the high-risk section of the AI Act — the part covering systems used in areas like hiring, education, critical infrastructure, law enforcement, and other settings where errors can seriously affect people’s rights or safety. Under the new(consilium.europa.eu)o August 2, 2028. (consilium.europa.eu) ### Why split the dates in two? Because the EU is trying to untangle two different buckets of AI. One bucket is stand-alone software — think systems used directly for scoring, screening, or decision support. The other is AI built into products already covered by sector-specific safety law(consilium.europa.eu)void companies dealing with overlapping compliance tracks at once. (consilium.europa.eu) ### What stays in place? The overall AI Act does. This is not a rewrite of the law’s core logic. The EU is still keeping its risk-based structure, where the stricter rules attach to the systems judged most dangerous. The Council and Parliament both framed the deal as streamlining, not gutting — and the text also restores a database-registration duty for providers that claim an exemption from high-risk classification. (consilium.europa.eu) ### What’s the deal with “nudifier” apps? That part got tougher, not softer. Parliament and Council agreed to ban AI systems that create child sexual abuse material or generate sexually explicit or intimate depictions of an identifiable person without that person’s consent. The ban covers putting such systems on the EU market, putting them into service, or using them. Parliament had pushed this point hard during the simplification talks. (europarl.europa.eu) ### Why was this happening now? Because the clock was about to become a problem. The AI Act entered into force on August 1, 2024, and applies in stages, with different obligations turning on over time. Some bans and general-purpose AI duties are already live, but the high-risk obligations were approaching fast. EU institutions argued that guidance, standards, and implementation plumbing were not ready enough for companies to comply cleanly by the original dates. (consilium.europa.eu) ### Why are critics unhappy? Because delays change power. If you’re a company building or deploying high-risk AI, more time means lower immediate compliance costs and less legal uncertainty. But if you’re worried about weak oversight, the same delay looks like a year-plus extension for systems that can still shape hiring, schooling, policing, or access to services before (consilium.europa.eu)is a practical fix or a softer word for postponement. (consilium.europa.eu) ### Does this mean the law is final? Not quite. The deal is provisional, which means it still needs formal approval by the Parliament and the Council before it becomes law. But politically, the hard part is done. Once both sides have agreed on dates and scope, the odds usually favor final sign-off. (consilium.europa.eu) ### Bottom line? The EU just chose a slower landing for the toughest part of its AI law. Companies get breathing room. Regulators get time to finish the missing guidance. But Europe also made a very clear trade — less immediate friction now, in exchange for slower enforcement on some of the systems it says matter most.