EU Council and Parliament fail to agree on AI Act changes after 12 hours of talks

The European Union has stalled its own AI reform again: after 12 hours of talks, EU member states and the European Parliament could not agree on amendments to the AI Act. The decisive attempt has now been pushed to May 2026, and the timelines that developers and providers of high-risk AI systems were counting on are left hanging with it.

Why the talks collapsed

The trilogue negotiations took place on Tuesday, 28 April 2026, and ended without a political agreement late at night. On the table was the so-called Digital Omnibus on AI, a package of amendments meant to simplify the application of the AI Act and reduce the regulatory burden on business.

For Brussels, this is not a technical tweak but a political test: the European Commission is trying both to keep its reputation as the author of the world's strictest AI rules and to show companies that Europe can respond to complaints about overly heavy compliance.

The problem is that, under the banner of "simplification," the sides are arguing about the very architecture of the law. Some see the package as a way to remove duplicated requirements and give companies more time; others see it as a rollback before the key provisions of the AI Act have even started to apply in full.

After the talks failed, the Cypriot presidency of the Council of the EU acknowledged only one thing: it had not been possible to reach agreement with the European Parliament. The next round is now expected in May, and the sides have less and less room to maneuver.

What the main dispute is about

The sharpest question concerns high-risk AI systems embedded in products that are already governed by sector-specific safety rules. The issue is whether such systems should be subject only to those sectoral rules, or whether the AI Act's special requirements should apply to them as well.

The European Parliament, backed by part of the industry, is pushing for a broader exemption, while the Council of the EU clearly does not want too many such systems removed from the scope of the general law. The talks got stuck at exactly this fork: it is no longer a dispute over wording, but over where product safety ends and standalone AI regulation begins.

• medical devices
• toys
• connected cars
• industrial equipment

Critics of the package argue that such relaxations will weaken protection of users' rights and safety. More than 40 civil society and human rights organizations had earlier warned that reworking the AI Act could hit oversight of biometrics, medical AI and systems affecting fundamental rights. Against this backdrop, the reaction of Dutch MEP Kim van Sparrentak was especially harsh:

"Big tech corporations are already opening the champagne, while

European companies that prepared for safety requirements are getting regulatory chaos."

Deadlines and risks

For the market, the calendar matters most right now. Under the current text of the AI Act, the key obligations for high-risk systems are due to start applying on 2 August 2026. The Omnibus was designed precisely to postpone that moment: to 2 December 2027 for standalone high-risk systems and to 2 August 2028 for systems embedded in regulated products.

But until the new version is agreed, voted on and officially published, the old deadline is the one legally in force. If the May talks drag on again, businesses could find themselves at August without the promised delay.

Not the whole package meets equal resistance, however. It contains one measure on which the sides had already reached notable consensus: a ban on AI systems that create intimate images without consent, including so-called nudification and child sexual abuse material. Yet even such a nearly uncontroversial provision did not save the deal. This shows the scale of the conflict: the dispute is not over a minor amendment, but over how far Europe is willing to go in softening its own AI regime in the name of competitiveness.

What this means

For the European AI market, the signal is fairly direct: do not count on looser rules in advance. Until the deal is closed, it is safer for companies to plan as if the 2 August 2026 deadline remains in force.

And for Europe itself, this failure is a reminder that its status as a global regulator rests not on loud laws, but on the ability to bring them to working condition on time and without chaos.