EU softens AI Act rules
- European Parliament negotiators and EU member states reached a provisional deal on May 7 to rewrite parts of the AI Act, easing compliance and delaying key duties. - The biggest change is timing: obligations for many high-risk AI systems move to December 2027, while the deal also adds an explicit EU ban on nudifier apps. - It matters because Brussels is shifting from “set the global rulebook” to “make the rules livable” as industry pressure and competitiveness fears reshape tech policy.
The EU’s AI Act was supposed to be the world’s toughest AI rulebook. Now Brussels is trimming it back before some of the hardest parts even fully bite. European Parliament negotiators and EU member states struck a provisional deal on May 7 to simplify the law, delay some obligations, and add a fresh ban on AI “nudifier” tools that generate non-consensual sexualized imagery. The basic risk-based structure stays. But the message changed — Europe still wants to regulate AI, just with less friction for companies. (consilium.europa.eu) ### What actually changed? The deal is part of the EU’s broader “Omnibus” simplification push — basically a cleanup drive aimed at cutting overlap, paperwork, and compliance costs across digital laws. For the AI Act, that means streamlining how certain high-risk systems are assessed, clarifying where AI rules stop and machinery or product-safety rules start, and softening some recurring obligations for companies. The negotiators also agreed to postpone parts of the timetable for high-risk AI. (consilium.europa.eu) ### Why was timing such a big fight? Because the original rollout was running into a practical problem: the standards and guidance companies need in order to comply were not ready fast enough. That left businesses staring at legal duties without clear technical instructions for meeting them. The political fix was delay. One of the headline changes pushes key obligations for many high-risk AI systems back by more than a year, with multiple reports pointing to December 2027 as the new date. (politico.eu) ### What counts as “high-risk” here? This is the part of the AI Act that covers systems used in sensitive settings — things like employment, education, critical infrastructure, and some safety-related products. These systems are not banned outright, but they face the heaviest compliance load: risk management, documentation, human oversight, data governance, and conformity checks. That is why delaying this section matters so much more than tweaking a side rule. It affects the expensive core of the law. (consilium.europa.eu) ### Why add a nudifier ban now? Because this was one area where lawmakers wanted to show they were not just weakening the law. The deal explicitly bans AI systems designed to create child sexual abuse material or sexualized deepfake imagery of identifiable people without consent — the so-called nudifier apps. That gives the package a sharper public-safety edge, and politically it helps offset the optics of delay. (europarl.europa.eu) ### So is this a rollback? Basically, yes — though not a full retreat. The core architecture of the AI Act survives, and Brussels is still treating AI risk as something that needs hard rules. But this is still a real softening. Politico called it the first major rollback or delay in the EU’s digital rulebook, and rights groups have argued the simplification drive is giving industry more room at the expense of stricter safeguards. (politico.eu) ### Why is Europe doing this now? Competitiveness. That is the short answer. EU capitals, companies, and the Commission have been under pressure to prove Europe can regulate AI without making itself a worse place to build AI products. Ursula von der Leyen backed the package as more innovation-friendly, while business-facing summaries of the deal highlight lower recurring administrative costs. Turns out the EU is (politico.eu)hate most. (europesays.com) ### What happens next? The deal is provisional, so it still needs the formal sign-off steps that turn a political agreement into final law. But the direction is now pretty clear. Europe is not abandoning the AI Act. It is rewriting the implementation story — slower on the hardest compliance burdens, tougher on obviously abusive uses, and more willing to admit that writing a rule is easier than making it workable. (consilium.europa.eu)ficial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/)) ### Bottom line The EU is still trying to be the place that sets AI rules first. But this week’s deal shows a different instinct taking over — not maximalism, but manageability. If the original AI Act was Brussels saying “we’ll draw the red lines,” this rewrite is Brussels admitting the map was too messy to use.
