EU AI Act Omnibus Trilogue Collapses After 12-Hour Brussels Session

The European Parliament, the Council of the European Union and the Commission ended a 12-hour trilogue in Brussels in the early hours of 29 April 2026 without agreement on the AI Act Omnibus reform. It was the second and final scheduled session before the August enforcement deadline. Talks resume next month.

What was supposed to be agreed

The Digital Omnibus, introduced by the Commission in November 2025, aims to streamline overlapping EU digital rules — AI Act, GDPR, Data Act, NIS2, Cyber Resilience Act — and align their compliance architectures. The headline AI Act change is treatment of high-risk obligations: the proposal links the effective date of compliance to the availability of harmonised standards and support tools, with long-stop dates of 2 December 2027 for standalone Annex III high-risk systems and 2 August 2028 for AI embedded in already-regulated products.

Where it broke

On Annex I — the list of EU sectoral safety legislation that already governs products into which AI is increasingly embedded. Medical devices, industrial machinery, toys, connected cars. The European Parliament wants AI inside these products to be governed solely by the sectoral rules, on the grounds that double regulation produces friction without proportionate safety gain. The Council and Commission resist, worried that exempting embedded AI from horizontal AI Act obligations creates a structural carve-out that erodes the framework.

MEP and AI Omnibus rapporteur Michael McNamara cautioned that routing governance through sectoral legislation could prove "deregulatory rather than simplifying" — language that captures Parliament's internal split between members who want lighter compliance and members who view simplification as a Trojan horse for weakening the AI Act outright.

What still applies on 2 August 2026

Whatever happens with the Omnibus, two things go live on 2 August 2026. First, Article 50 transparency obligations for newly launched generative AI systems — disclosure and content-marking capabilities. Second, the requirement for each EU member state to establish at least one AI regulatory sandbox at national level. Luxembourg, with the CSSF, the Ministry of Digitalisation and the Luxembourg Institute of Science and Technology in the mix, has been preparing its sandbox architecture for months.

What is next

A resumed trilogue in May, then likely an agreement that splits the Annex I question — partial sectoral primacy for some product categories, AI Act primacy retained for others. Industry will read whatever lands carefully. So will the German, French and Italian governments whose industrial bases (cars, machinery, medical devices) are most affected. The political timetable is tight, the substance is technical, and the cost of getting it wrong shows up not in 2026 but in three years' time when the long-stop dates start binding.