Council and Parliament agree changes to EU AI Act

Negotiators from the Council Presidency and the European Parliament have reached a provisional agreement to amend the EU AI Act under the Omnibus VII simplification package. The changes are designed to reduce compliance pressure on companies while keeping the core risk-based structure of the AI Act intact. The agreement largely follows the Commission’s proposal to delay certain obligations until technical standards and compliance tools are in place.

Under the compromise, the application of high‑risk AI rules will be postponed. Stand‑alone high‑risk AI systems would fall under the AI Act from 2 December 2027, while high‑risk AI systems embedded in regulated products would become subject to the rules from 2 August 2028. The deadline for national authorities to establish AI regulatory sandboxes is also delayed to 2 August 2027, giving regulators and businesses additional preparation time.

The package expands selected regulatory exemptions beyond SMEs to include small mid‑cap companies and simplifies compliance in limited scenarios. At the same time, lawmakers reinforced safeguards against harmful uses of AI by explicitly prohibiting systems that generate non‑consensual sexual or intimate material and child sexual abuse material. The grace period for implementing transparency tools for AI‑generated content is shortened to three months, with a new deadline of 2 December 2026.

Several accountability measures were restored during negotiations. Providers claiming exemption from high‑risk classification must still register their systems in the EU high‑risk AI database, and the strict necessity standard for processing sensitive data to address bias has been reinstated. The agreement also clarifies the EU AI Office’s supervisory role over certain general‑purpose AI systems, limits overlap with sectoral legislation, and introduces partial exemptions for machinery rules. Formal approval by the Council and Parliament is expected following legal and linguistic review.