EU talks fail on watered-down AI

Europe’s AI rulebook is back in limbo. After a marathon negotiating session in Brussels on April 28, EU member states and European Parliament negotiators failed to agree on a package that would have softened parts of the AI Act and pushed back some deadlines. That matters because the clock did not stop. For companies building or deploying AI in Europe, August is still the date to plan around — even if politicians are still arguing over whether the law is too strict. (iapp.org) ### What actually broke in the talks? The short version is that the Parliament and the Council could not agree on how far to go in trimming the law. Cyprus, which currently holds the rotating Council presidency, was negotiating for member states. After about 12 hours, the two sides walked away without a deal, and officials said they would t(iapp.org)he EU should now lighten it before some of the hardest obligations bite. (iapp.org) ### Why were they trying to soften it? Because the original law is colliding with implementation reality. Companies have been complaining that guidance arrived late, some technical standards are still incomplete, and the compliance burden is hard to map onto real products. Policymakers have also been trying to reduce overlap between the AI (iapp.org)mes. Turns out that is where part of the negotiation jammed up. Parliament and Council disagreed on how the AI Act should interact with those other frameworks. (iapp.org) ### Which deadline is spooking people? The immediate pressure point is August 2. The Commission’s own materials make clear that obligations for general-purpose AI model providers started applying on August 2, 2025, and the next major compliance step lands on August 2, 2026, including rules tied to certain high-risk systems. Because this wee(iapp.org)ritten timetable still stands. (digital-strategy.ec.europa.eu) ### Why does “high-risk” matter so much? Because that is the part of the AI Act that moves beyond broad transparency and into product-style compliance. High-risk systems can trigger requirements around risk management, documentation, human oversight, data governance, monitoring, and conformity assessments. In plain English — this(digital-strategy.ec.europa.eu)dor contracts, testing, and audit trails. That is why even a possible delay changes budgets right away. (iapp.org) ### So are companies pausing or pushing ahead? Mostly pushing ahead. The logic is simple — if you wait for Brussels to settle an argument and the delay never comes, you are late. Governance teams are still doing audits, mapping systems, and sorting which tools might fall into the high-risk bucket. The weird part is that the law is supposed to creat(iapp.org)er one timetable while lobbying for another. (iapp.org) ### Why is this politically awkward for the EU? Because Europe sold the AI Act as the world’s flagship AI law. Now, before some of its toughest provisions fully land, the bloc is already debating simplification. That does not mean the law is collapsing. But it does show the tension at the heart of EU tech policy — Brussels wants to lead on (iapp.org)ent and investment. (iapp.org) ### What happens next? Negotiators are expected to reconvene in May. Until then, the safest assumption for companies is boring but clear — plan for the current law, not the hoped-for rewrite. If a softer deal eventually emerges, firms can scale back. If it does not, they will at least not be scrambling in late July. (iapp.org)as not just another Brussels process story. It was a reminder that Europe’s AI law is now in the painful phase where grand principles turn into checklists, audits, and legal exposure — and that is exactly where consensus gets harder.