Source: Lewis Silkin

On 7 May 2026 the Council presidency and European Parliament struck a provisional deal to slim down parts of the EU AI Act under Omnibus VII. The core change is a timetable delay: standalone high‑risk AI systems will be subject to obligations from 2 December 2027, and high‑risk AI systems embedded in products from 2 August 2028. The agreement keeps the Act’s overall architecture while introducing several targeted amendments rather than a full rewrite.

Key substantive amendments include a new prohibition on using AI to generate non‑consensual sexual content (so‑called nudification tools) and AI‑generated child sexual abuse material (CSAM). The standard of "strict necessity" for processing special categories of personal data for bias detection has been reinstated. Providers must register AI systems in the EU database even where they consider their system exempt from high‑risk classification. SME exemptions now extend to small mid‑caps (up to 500 employees), the deadline for national AI regulatory sandboxes is moved to 2 August 2027, transparency measures for AI‑generated content must be implemented by 2 December 2026, and the Commission is empowered to adopt delegated acts to add health and safety requirements under the Machinery Regulation, which receives a full exemption from direct AI Act applicability. The deal clarifies the AI Office’s supervisory competence for systems built on general‑purpose models developed by the same provider while reserving national authority control in law enforcement, border management, judicial and financial domains. The provisional agreement requires endorsement by both institutions and a legal‑linguistic scrub before formal adoption, expected within weeks. For businesses, the extra 16 months (and 12 months for embedded AI) is presented as breathing space for more robust compliance programmes, and the analysis frames the outcome as pragmatic simplification rather than substantive reform.