The EU institutions have reached a preliminary political agreement on the AI Act Omnibus, following intensive trilogue negotiations and a failed “final” negotiating round last week. The agreement is not yet final. It must still undergo review and formal adoption by the Council and Parliament before entering into force. However, the political deal provides the clearest indication to date of the changes expected to be incorporated into the AI Act.
Key changes to the AI Act
New compliance deadlines for high-risk AI systems
- 2 December 2027 for standalone high-risk AI systems under Annex III.
- 2 August 2028 for high-risk AI systems integrated into regulated products (Annex I).
New implementation deadline for AI transparency obligations
- 2 December 2026 for Article 50 transparency and digital watermarking requirements relating to AI-generated content.
Industrial AI and sector legislation
- The Commission will be able to limit the application of the AI Act where sector-specific legislation already contains equivalent AI requirements, including in sectors such as medical devices and machinery.
New prohibited AI practices
- AI systems generating non-consensual sexual or intimate content and child sexual abuse material will be prohibited in the EU.
Machinery Regulation
- The Machinery Regulation (EU) 2023/1230 will no longer fall within the AI Act’s direct scope. Instead, the Commission will be empowered to adopt delegated acts under the Machinery Regulation introducing AI-related health and safety requirements for relevant high-risk AI systems.
Additionally, two proposed relaxations appear not to have been retained in the preliminary political agreement. Providers will still be required to register certain AI systems in the EU high-risk database where they consider those systems exempt from high-risk classification. The agreement also retains the “strict necessity” threshold for processing special category personal data for bias detection and correction.
Click here for the press release.
