The European Parliament and Council have reached provisional agreement on the AI Omnibus, which was introduced by the European Commission in November 2025 and contains proposals to amend the EU AI Act. The key points are:
High-risk AI: extension of compliance deadlines
The AI Act’s core obligations on providers and deployers of high-risk AI systems were due to become applicable in August 2026. The Parliament and Council have agreed to extend this deadline as follows:
- 2 December 2027 for standalone high-risk AI systems; and
- 2 August 2028 for high-risk AI systems embedded in products.
AI transparency: extension of grace period
The AI Act contains transparency or “watermarking” obligations for AI systems placed on the market before 2 August 2026, which were due to become applicable on that date. This deadline has been delayed until 2 December 2026.
Use of special category data for bias detection and correction
The Commission proposed extending the right to process special category data (as defined in the GDPR) for the purpose of ensuring bias detection and correction (which is currently limited to providers of high-risk AI systems where it is strictly necessary) to deployers of high-risk systems and models and providers and deployers of other types of AI systems. The Council and Parliament have agreed to reinstate the requirement that special category data may only be processed for this purpose where it is strictly necessary.
Ban on explicit deepfakes
The Council and Parliament have agreed to introduce a ban on AI systems that either have the purpose of generating non-consensual intimate imagery and child sexual abuse material or do not have in place reasonable safety measures to prevent this. The prohibition also applies to deployers using AI systems to create such content. This addresses a regulatory gap highlighted by recent scandals in which AI models produced sexualised images of real people, including minors, as a product feature. The ban is due to take effect on 2 December 2026.
Overlap between AI Act and sectoral legislation
The Council and Parliament have agreed to tackle overlapping requirements of the AI Act and sector-specific laws. Machinery is excluded from the direct application of the AI Act in cases where there is overlap, with health and safety matters being addressed directly under the Machinery Regulation. For other sector-specific laws, e.g. those applicable to medical devices, connected vehicles and toys, the Commission will issue implementing acts.
Registration of high-risk AI systems in the EU database
The Commission proposed an exemption from the obligation for providers to register AI systems in the database for high-risk AI systems where the system does not pose a significant risk of harm to health, safety or fundamental rights. The Council and Parliament have agreed to reinstate the registration obligation.
Expanded exemptions for small mid-cap entities
The Council and Parliament have agreed with the Commission’s proposal to extend some of the exemptions from the AI Act’s obligations currently applicable to SMEs to SMCs (small mid-cap entities).
Allocation of responsibilities between EU AI Office and national authorities
The provisional agreement clarifies the respective competences of the AI Office and national authorities for the supervision of AI systems based on general-purpose AI models.
AI literacy
The press releases published by the Council and Parliament do not address the issue of AI literacy. The Commission proposed moving the duty for AI providers and deployers to ensure AI literacy among their staff (which became applicable in February 2025) from the organisation itself to the EU and national authorities. In the drafts previously published, both institutions took slightly different positions on this issue, so it will be interesting to see their agreed position, once this is made available.
