AI Omnibus Deal: EU Pushes High-Risk AI Rules to December 2027

The European Parliament and the Council of the European Union reached a provisional political agreement on the so-called AI Act Omnibus on Thursday 7 May 2026 — a package of targeted amendments designed to streamline implementation of the EU’s landmark Artificial Intelligence Act while introducing new safeguards against the most harmful forms of generative misuse.

The agreement, reached at trilogue level between the Council, the Parliament and the European Commission, forms part of the broader Digital Omnibus initiative launched in November 2025. The Commission, which proposed the simplifications in response to industry concerns about implementation readiness, framed the outcome as an exercise in pragmatic adjustment rather than substantive retreat from the AI Act’s core architecture.

Three structural changes

The provisional agreement makes three changes of structural importance. First, the application date for obligations on high-risk AI systems (HRAIS) — originally set for 2 August 2026 — is pushed to 2 December 2027 for standalone high-risk systems, and 2 August 2028 for high-risk AI embedded in regulated products such as medical devices, toys and machinery. The Commission had previously argued that the necessary harmonised standards would not be available in time for the original deadline.

Second, the watermarking obligations under Article 50(2) — requiring providers of generative AI systems to mark synthetic content in a machine-readable way — are postponed from 2 August 2026 to 2 December 2026, with a grandfathering clause for systems already on the market before the original date.

Third, the agreement introduces a new prohibition, effective 2 December 2026, on AI systems designed to generate non-consensual sexually explicit or intimate imagery, including so-called “nudifier” applications, and on systems that generate child sexual abuse material. Violations may trigger fines of up to €35 million or 7% of annual worldwide turnover, whichever is higher.

Cyprus presidency: a question of competitiveness

The agreement was announced during the Cyprus Presidency of the Council. Marilena Raouna, Cyprus’s Deputy Minister for European Affairs and the Council representative at the trilogue, framed the outcome in terms of strategic competitiveness: “Today’s agreement on the AI Act significantly supports our companies by reducing recurring administrative costs. It ensures legal certainty and a smoother and more harmonised implementation of the rules across the Union, strengthening EU’s digital sovereignty and overall competitiveness.”

Raouna added: “At the same time, we are stepping up the protection of children targeting risks linked to the AI systems. This agreement is clear evidence of our institutions’ ability to act swiftly and deliver on our commitments. It marks the first deliverable under the ‘One Europe, One Market’ roadmap agreed by the three institutions, well within the set deadline.”

What stays unchanged

Several core provisions of the AI Act are explicitly preserved. The prohibitions under Article 5 — covering social scoring, subliminal manipulation, exploitation of vulnerabilities, and real-time biometric remote identification in public spaces — remain in force from their original date of 2 February 2025. The governance rules and obligations for general-purpose AI models, which became applicable on 2 August 2025, are also undisturbed. Most transparency obligations under Article 50 — including the requirement to disclose AI interaction, the emotion recognition and biometric categorisation duties, and deepfake labelling — will still start on 2 August 2026.

The Commission published on 8 May 2026, the day after the political agreement, its draft guidelines on AI transparency obligations under Article 50. The targeted consultation on those guidelines runs until 3 June 2026 and is expected to produce the final implementing guidance in the autumn.

The road to formal adoption

The provisional agreement must still be formally adopted by both co-legislators before it becomes law. Given the proximity of the original 2 August 2026 deadline, the legislative timetable is being compressed: formal adoption is expected by July 2026, with publication in the Official Journal of the European Union immediately thereafter. For companies developing, deploying, or distributing AI systems in the EU, the immediate priority remains scoping assessment and governance framework development. The extended deadlines do not alter the underlying compliance obligations; they redistribute them across time.