EU fails deal on AI rules

Europe’s AI rulebook just hit a very European kind of snag — the law exists, the deadline is close, and lawmakers still can’t agree on how much to soften it before the hard parts kick in. After 12 hours of negotiations in Brussels on April 29, EU member states and European Parliament negotiators walked away without a deal on proposed changes to the AI Act. That matters because the next big compliance date is August 2, 2026, and without a fix, the stricter timeline stays in place for now. (msn.com) ### What were they trying to change? The fight was over an “omnibus” package — basically a cleanup bill meant to simplify parts of the AI Act before more obligations start applying. The original law entered into force in 2024, but its requirements phase in over time. This year’(msn.com)and compliance machinery were fully ready. (economictimes.indiatimes.com) ### What broke the talks? The sharpest dispute was about overlap with sector-specific regulation. Parliament’s center-right, backed by Germany, pushed for products such as medical devices and industrial machinery to comply thro(economictimes.indiatimes.com) paperwork manufacturers face. Negotiators couldn’t close that gap. (politico.eu) ### Why does August matter so much? Because August 2 is not a symbolic date. It is when major obligations for high-risk AI systems are due to start applying under the current law if nothing changes. Both Parliament and the Council had been working around that clock, and earlier compromise ideas would have(politico.eu)eans those relief valves are not law. Yet. (computerworld.com) ### What counts as “high-risk” here? This is the part that makes the story feel bigger than Brussels process drama. High-risk AI in the EU includes systems used in sensitive areas like employment, education, critical infrastructure, and certain products already c(computerworld.com)icants, supports classroom decisions, or sits inside regulated equipment. (matheson.com) ### Why are schools and other users still stuck? Because legal text drives procurement, compliance, and internal policy. If a school system, hospital, or manufacturer does not know whether the August obligations will be delayed, narrowed, or rerouted through another rulebook, it has to plan for several ve(matheson.com)d a bias toward human sign-off until the law settles. That last point is an inference from the unresolved timeline and scope debate, but it follows pretty directly from the compliance uncertainty. (iapp.org) ### So is the AI Act falling apart? No — and that is the important distinction. The AI Act itself is still in force. What failed was a late effort to simplify and delay parts of its implementation. In other words, Europe did not scrap its AI law. It failed to agree on a softer landing for the next phase. (economi([iapp.org)wmakers-fail-to-reach-deal-on-watered-down-ai-rules/articleshow/130593132.cms)) ### What happens next? Talks are expected to resume in May, which leaves a very small window to strike a political deal, finish the legal text, and get it formally adopted before August. If that slips, companies may have to prepare for the original timetable while hoping lawmakers eventually grant relief after the fact — basically the worst of both worlds. (thenextweb.com) ### Bottom line? This was supposed to be the tidy-up round. Instead, it exposed a deeper argument about whether Europe wants one AI rulebook layered on top of everything else, or a looser system that lets sector regulators take the lead. Until that gets resolved, the law is real, the deadline is real, and the uncertainty is real too. (politico.eu)