EU AI Omnibus Deal Reached: More clarity on AI Act obligations
In the early hours of 7 May 2026, after roughly nine hours of negotiations, the European Parliament, Council, and Commission reached political agreement on the AI Omnibus during the third political trilogue meeting. The deal, struck after the previous trilogue collapsed on 28 April over the question of how the AI Act should interact with sectoral product safety legislation, brings to a close several months of intense negotiations and clears the path for postponement of the AI Act's high-risk obligations before the original 2 August 2026 deadline takes effect.
What is the AI Omnibus?
The AI Omnibus is a legislative proposal published by the European Commission on 19 November 2025, forming part of the broader Digital Simplification Package. Its goal is to amend the EU AI Act to simplify certain obligations and, most importantly for industry, postpone the application of high-risk AI compliance requirements that would otherwise have applied from August 2026. The package was prompted by widespread acknowledgement that the supporting ecosystem (harmonised standards, conformity assessment infrastructure, regulatory guidance, and notified body capacity) would not be in place in time for the original deadline.
High-risk AI deadlines: delays confirmed
The headline element of the deal is the confirmation of postponement dates for the AI Act's high-risk obligations, an outcome that had already been broadly agreed by Council and Parliament before the trilogue process began:
Annex III high-risk AI systems (those used in areas such as employment, education, credit scoring, and law enforcement): obligations now apply from 2 December 2027.
Annex I high-risk AI systems (AI embedded in products covered by EU sectoral safety legislation, such as machinery and medical devices): obligations now apply from 2 August 2028.
These fixed dates replace the conditional mechanism originally proposed by the Commission, providing greater legal certainty for providers and deployers building their compliance programmes.
What is no longer directly covered by the EU AI Act
The most contested element of the trilogue, and the issue that caused the 28 April collapse, were the contents of Annex I, specifically whether AI systems already covered by EU sectoral product safety legislation should remain under the AI Act's direct scope (Section A) or be governed primarily by their sectoral frameworks (Section B). The Parliament had pushed for a transfer of all Section A products to Section B which the Council initially resisted.
The agreed compromise lands somewhere in the middle:
The Machinery Regulation has been moved from Section A to Section B of Annex I. AI systems falling under the Machinery Regulation will therefore be addressed primarily through that sectoral framework rather than through the AI Act's direct scope.
The Medical Devices Regulation (MDR), In Vitro Diagnostics Regulation (IVDR), and Radio Equipment Directive (RED) have not been moved. Despite Parliament negotiators raising the question early in the trilogue, these sectors remain in Section A of Annex I.
A new Article 2(13) mechanism allows the Commission to limit, via delegated implementing acts, the application of specific AI Act requirements where Section A sectoral legislation provides an equivalent or higher level of protection of health, safety, or fundamental rights.
The definition of 'safety component' in Article 3(14) has been tightened.
This sectoral architecture compromise is the single most consequential outcome for industrial AI providers operating across the EU.
New prohibitions on non-consensual intimate imagery
The deal introduces new prohibitions in the AI Act's Article 5 on prohibited practices:
A ban on AI systems generating non-consensual intimate imagery (so-called 'nudifiers').
A ban on AI systems generating child sexual abuse material (CSAM).
These prohibitions, which were not in the Commission's original proposal but were added during the legislative process with strong cross-party support, will apply from 2 February 2027, giving providers a transition period to ensure compliance. The agreed text also clarifies that the prohibitions are without prejudice to remedies available under national laws for individuals to protect their fundamental rights.
Transition periods narrowed
Several transition periods sought by industry were narrowed during the negotiations:
The grace period for Article 50(2) transparency and labelling obligations covering generative AI synthetic content has been set at 4 months, with compliance required by 2 December 2026. This is shorter than the 6-month period favoured by Council and Commission, and below the 12-month period requested by industry, but represents a midpoint between the Parliament and Council positions.
Member States have been given additional time to establish national AI regulatory sandboxes, with the deadline pushed back from the original 2 August 2026.
Next steps?
The political agreement reached on 7 May is not yet the final text. A technical meeting is expected before the end of this week to finalise the wording, particularly within the recitals. The Parliament and Council will then need to formally adopt the consolidated text, with publication in the Official Journal of the European Union (OJEU) anticipated within roughly two months. This timeline should allow the postponement of the high-risk AI obligations to take legal effect before the original 2 August 2026 deadline.
Until publication, the original AI Act timelines remain legally in force.
