Brussels talks collapse; EU keeps Aug. 2 AI transparency deadline
- EU negotiators failed on April 28 to revise the AI Act through the Digital Omnibus, so the bloc’s original August 2, 2026 compliance dates remain alive. - The key split was over exempting AI inside regulated products like medical devices and machinery; Parliament pushed, but the Council and Commission refused. - For model makers, that keeps pressure on Article 50 labeling and GPAI duties already phased in since August 2, 2025.
The EU AI Act is now back in its hardest mode — the original one. After 12 hours of talks in Brussels on April 28, negotiators failed to agree on changes that could have pushed parts of the law back. So the practical message for companies is simple: stop planning around a delay that does not exist yet. The August 2, 2026 deadline is still the live one for the next big wave of obligations. (iapp.org) ### What actually collapsed in Brussels? This was a fight over the AI Act’s place inside the EU’s broader “Digital Omnibus” simplification push. The Commission, Parliament, and Council were trying to settle targeted changes before tougher requirements start applying in 2026. They did not get there. The April 28 trilogue was supposed to be the second and final negotiating session, but it broke down instead of producing a deal. (iapp.org) ### Why did the talks fail? The biggest sticking point was AI that sits inside already regulated products — medical devices, industrial machinery, connected cars, toys, that kind of thing. Parliament pushed for a carve-out so some of those systems would rely more on sector-specific rules instead of taking the full extra load of AI Act requirements. The Coun(iapp.org)ion. (iapp.org) ### So what deadline is still real? August 2, 2026. That is when the Commission’s enforcement powers for general-purpose AI model obligations kick in, and it is also when Article 50 transparency obligations apply. Those Article 50 rules cover things like telling people when they are interacting with AI and making AI-generated or AI-manipulated content detec(iapp.org)le Brussels argued. (digital-strategy.ec.europa.eu) ### Wait — didn’t some AI rules already start? Yes. This is the part that trips people up. The obligations for providers of general-purpose AI models entered into application on August 2, 2025. That means the compliance story is phased, not one big switch-flip in 2026. Providers of GPAI models placed on the market after August 2, 2025 are already inside the regime, while the Commission gets full enforcement powers from August 2, 2026. (digital-strategy.ec.europa.eu) ### Who gets extra time until 2027? Providers of general-purpose AI models that were placed on the market before August 2, 2025 get until August 2, 2027 to comply. That grandfathering point matters a lot because it splits the market into older models with a longer runway and newer models facing the stricter timetable. So when people say “(digital-strategy.ec.europa.eu)arted in 2025. (digital-strategy.ec.europa.eu) ### What does Article 50 mean in practice? It is the transparency layer. If a system interacts directly with people, users may need to be told they are dealing with AI. If a system generates or manipulates text, images, audio, or video, outputs may need to be marked so they are detectable as artificial. There are exceptions — standard edi(digital-strategy.ec.europa.eu)n AI is out, labeling is in. (ai-act-service-desk.ec.europa.eu) ### Why are companies so jumpy about this? Because uncertainty is already changing behavior before full enforcement arrives. OpenEvidence, a clinical AI platform with broad U.S. hospital usage, pulled out of the UK and EU this week and blamed regulatory uncertainty around the AI Act. One company leaving does not prove the whole market is freezing. B(ai-act-service-desk.ec.europa.eu)ompliance lift right now. (iatrox.com) ### What’s the bottom line? The failed Brussels talks did not create clarity, but they did kill the easiest excuse for waiting. Until lawmakers actually pass a change, August 2, 2026 is the date that matters — and companies building or deploying AI in Europe need to treat it that way. (iapp.org)
