Parliament and Council aim to reach a swift compromise.

In November 2025, the European Commission presented a proposal for a Digital Omnibus in the field of artificial intelligence (AI). Immediately following the adoption of the European Parliament’s negotiating mandate on 26 March, trilogue negotiations between the Council of the European Union (EU) and the Parliament began. The Council had already adopted its negotiating mandate on 13 March. While some points are largely uncontroversial, disagreement remains particularly regarding the treatment of industrial AI.

Postponement of deadlines uncontroversial

The European Commission had proposed linking the timing of the entry into application of the provisions on high-risk AI to the availability of supporting tools, including the necessary standards, and postponing it accordingly. However, the rules were to become applicable after 16 months at the latest. Both Parliament and Council oppose a flexible mechanism and instead favour fixed deadlines for the application of the high-risk AI provisions – 2 December 2027 for stand-alone high-risk AI systems under Annex III and 2 August 2028 for high-risk AI systems embedded in products under Annex I.

Furthermore, there is agreement between the two institutions that providers of AI systems under Annex III should continue to be required to register these systems in the EU database for high-risk systems. The European Commission had proposed abolishing this registration obligation if providers conclude, on the basis of a documented assessment, that the system is not high-risk.

Industrial AI as main point of contention

The main disagreement concerns the treatment of industrial AI, i.e. AI systems under Annex I that form part of already regulated products. These include, for example, machinery, protective equipment or medical devices. The European Parliament is calling for AI systems – contrary to the current provisions of the AI Act – not to be classified as high-risk solely because they are used in regulated products and could theoretically be safety-relevant. Instead, a purpose-based assessment should be carried out, with AI only being classified as high-risk where it is actually considered safety-relevant. The Council rejects this approach due to concerns about inconsistencies and duplication of obligations between the AI Act and sectoral legislation.

In addition to the treatment of industrial AI, two further lines of conflict shape the negotiations: the institutional role of the AI Office and the scope of fundamental rights protection. While Parliament is seeking stronger central enforcement powers at EU level, the Council emphasises national competences and the principle of subsidiarity. As regards fundamental rights protection, the objectives are broadly aligned, for example with regard to a ban on nudifier apps, bias rules and data protection, but the exact scope of obligations remains contested.

Outlook

The negotiations are under considerable pressure as the provisions on high-risk AI systems under the AI Act will become applicable in August this year. The aim of the trilogues, which were held several times a week, was therefore to reach an agreement by the end of April. However, the most recent trilogue on 28 April ended without a result. A further trilogue is expected to take place in the coming weeks, although no specific date has yet been confirmed.