EU and Parliament Fail to Reach AI Act Omnibus Deal After 12-Hour Trilogue; Talks Set for May

EU officials and European Parliament lawmakers walked away from a 12‑hour trilogue on Tuesday without a deal on the AI Act omnibus package, the final scheduled political session before the package’s deadline. The talks, which also covered amendments to the GDPR, the e‑Privacy Directive and the Data Act, will resume in May, according to Reuters.

Cyprus, holding the rotating EU Council presidency, said a deal with the Parliament was simply not possible. The impasse reflects a deeper split over a single, contentious question: should high‑risk AI systems that are already embedded in products governed by existing EU safety rules – such as medical devices, toys, connected cars and industrial machinery – be exempt from the AI Act’s additional requirements?

The European Parliament, backed by industry groups, argues for a sectoral carve‑out, insisting that existing product‑safety legislation is sufficient. Member‑state representatives in the Council have shown limited enthusiasm for such a broad exemption, fearing it would undermine the AI Act’s protective framework.

Critics of the omnibus package, including a coalition of privacy and civil‑rights organisations, view the proposed changes as a rollback of hard‑won safeguards. More than 40 groups signed a letter in mid‑April warning that the amendments could weaken fundamental‑rights protections for biometric identification, AI in schools and medical AI. Researchers echo these concerns, warning that diluting the AI Act before its core provisions take effect could erode Europe’s reputation as a global standard‑setter.

Parliament’s lead negotiator on the AI omnibus, Michael McNamara, acknowledged the difficulty of overlapping rules but cautioned that shifting AI governance into sector‑specific laws could end up “deregulatory rather than simplifying.” The debate underscores the challenge of balancing regulatory certainty for businesses with the EU’s ambition to maintain a high‑level of AI oversight.

The urgency of the negotiations is structural. The AI Act’s core obligations for high‑risk AI systems are set to apply from August 2, 2026, just three months away. The omnibus aims to push that deadline to December 2, 2027, for stand‑alone high‑risk systems, and to August 2, 2028, for those embedded in regulated products. Achieving that postponement requires a final political agreement, a formal Parliament vote, Council endorsement and publication in the Official Journal within a matter of weeks.

If talks stall again in May and no agreement is reached before June, the original August 2026 deadline will stand. Companies that have planned for the extended timeline would face immediate compliance obligations for which many have not prepared – a scenario Brussels has been working hard to avoid.

One provision that did garner consensus is a ban on AI systems that generate non‑consensual intimate images, including child sexual‑abuse material. The measure was added after controversy surrounding Elon Musk’s Grok chatbot’s nudification capabilities in late 2025. That even this area of agreement could not keep the broader talks from collapsing highlights how intractable the sectoral exemption question remains.

The next round of talks in May will determine whether the EU can still claim an orderly approach to its ambitious AI regulation or whether the world’s most comprehensive AI rules will stumble just as their toughest provisions are set to bite.