EU and European Parliament fail to agree on AI Act amendments after 12 hours of negotiations

The EU has once again stalled its own AI reform: after 12 hours of negotiations, EU member states and the European Parliament failed to agree on amendments to the AI Act. The decisive attempt has now been postponed to May 2026, and with it, the deadlines that developers and suppliers of high-risk AI systems were counting on have also been put on hold.

Why the negotiations failed

The trilogue negotiations took place on Tuesday, April 28, 2026, and concluded without political agreement deep into the night. On the table was the so-called Digital Omnibus on AI — a package of amendments designed to simplify the application of the AI Act and reduce regulatory burden on business. For Brussels, this is not a technical tweak but a political test: the European Commission is simultaneously trying to maintain its reputation as the author of the world's strictest AI rules and show companies that Europe can respond to complaints about excessive compliance burden.

The problem is that under the guise of "simplification," the parties are actually debating the very architecture of the law. While some see in the package a way to eliminate duplicate requirements and give companies more time, others view it as a rollback even before the key provisions of the AI Act have begun operating at full force.

After the negotiations collapsed, Cyprus's EU Council presidency acknowledged only one thing: agreement with the European Parliament could not be reached. The next round is now expected in May, and time to maneuver is running out for both sides.

What the main dispute is about

The most acute question concerns high-risk AI systems embedded in products that are already regulated by sector-specific safety rules. The issue is whether such systems should be subject only to the relevant sectoral regulations, or whether special requirements from the AI Act should additionally apply to them. The European Parliament, with support from parts of the industry, is pushing for a broader exemption, while the EU Council clearly does not want too large a withdrawal of such systems from the scope of the general law. This is the fork where negotiations got stuck: it is no longer a dispute about wording, but about where product safety ends and independent AI regulation begins.

• medical devices
• toys
• connected cars
• industrial equipment

Critics of the package argue that such concessions would weaken the protection of user rights and safety. More than 40 civil society organizations and human rights groups previously warned that a reassembly of the AI Act could undermine control over biometrics, medical AI, and systems affecting fundamental rights. Against this backdrop, the reaction of Dutch MEP Kim van Sparrentak was particularly harsh:

"Major tech corporations are already uncorking champagne, while

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

Deadlines and risks

Right now, the calendar is most important for the market. Under the current version of the AI Act, key obligations for high-risk systems are supposed to begin applying on August 2, 2026. The Omnibus was designed precisely to defer this moment: to December 2, 2027 for standalone high-risk systems and to August 2, 2028 for systems embedded in regulated products. But as long as the new version is not agreed upon, voted on, and officially published, legally it is the old deadline that holds. If the May negotiations drag on again, business could find itself in August without the promised extension.

However, not every part of the package encounters the same resistance. It includes one measure on which the parties already had significant consensus: a ban on AI systems creating intimate images without consent, including so-called nudification and child sexual abuse material. But even the presence of such an almost undisputed rule did not save the deal. This shows well the scale of the conflict: the dispute is not about an individual amendment, but about how far Europe is willing to go in softening its own AI regime for the sake of competitiveness.

What this means

For the European AI market, the signal is quite direct: companies cannot count on regulatory relief in advance. Until the deal is closed, companies are safer planning their operations as if the August 2, 2026 deadline remains in force. And for Europe itself, this failure is a reminder that the status of global regulator rests not on loud laws, but on the ability to bring them to working condition in a timely manner and without chaos.