EU governments and MEPs strike provisional pact diluting key AI Act rules

Europe’s AI rulebook is getting a rewrite before its toughest deadlines fully bite. That matters because the EU spent years selling the AI Act as the world’s strictest framework, and now it is easing parts of it in the name of competitiveness. On Thursday, May 7, EU governments and European Parliament negotiators reached a provisional deal to simplify some obligations and push back parts of the timetable. The move does not kill the AI Act. But it does show Europe blinking. ### What changed today? The Council presidency and Parliament negotiators agreed on a provisional package to “simplify and streamline” parts of the AI Act. This sits inside the EU’s broader Omnibus VII simplification agenda — basically a cleanup effort meant to cut compliance burdens across digital laws. The agreement is political for now, not final. Member states and the European Parliament still have to sign it off. ### Why is the EU softening its own law? Because the political mood changed. When the AI Act was passed, the center of gravity was safety, rights, and setting a global standard. Now the pressure point is industrial competitiveness — especially the fear that European competitive costs for companies. ### Which rules are being diluted? The public description is broad, not clause-by-clause. The core idea is lighter administration and slower implementation for some obligations, especially the pieces that were about to come due on a tight schedule. Legal briefings tracking this is not a philosophical rewrite of the whole law. It is a targeted relief package before enforcement ramps up. ### Why does timing matter so much? Because EU tech groups have been arguing that they were being asked to comply while key guidance was still unsettled. One flashpoint has been the code of practice for general-purpose AI models — the practical playbook for how model providers could to ship products. That timing problem made delay almost as important as deregulation. ### Who pushed for this? A very recognizable group of European industrial and tech chiefs. In a joint public appeal published this week, leaders from ASML, Airbus, Ericsson, Mistral AI, Nokia, SAP, and Siemens called for faster action to protect Europe’s technology competitiveness. That does not mean absolutely all lining up on the same side. ### Does this mean the AI Act is falling apart? Not really. The AI Act’s risk-based structure is still there. The EU is not abandoning the idea that higher-risk systems should face stricter obligations. What is changing is the implementation posture. Think of it less like tearing up the map and more like moving checkpoints farther doing that being first with rules can also mean being first with friction. ### What still has to happen? Formal endorsement. Provisional deals in Brussels are real progress, but they are not the last word. The Council and Parliament both still need to complete the approval process. Until that happens, companies have a strong signal about direction, not absolute legal closure. ### Bottom line The EU is still trying to regulate AI aggressively. But now it is trying to regulate it without kneecapping its own industry. That is the real news here — not that Europe gave up, but that competitiveness has moved to the front of the room.