Two institutional positions arrived in one week. On 13 March 2026 the Council agreed its position on streamlining the AI Act. On 16 March a joint parliamentary committee position was adopted by 101 votes to 9, with 8 abstentions. A plenary vote was expected on 26 March, after which Parliament and Council begin negotiating a common text.
What moves
Both positions push the application date for high-risk AI systems to 2 December 2027, and to 2 August 2028 for systems covered by EU sectoral safety legislation. The establishment of regulatory sandboxes is postponed to 2 December 2027 as well. Watermarking obligations, by contrast, are proposed on a shorter clock than the rest of the package.
What moves the other way
The prohibitions expanded. Both positions back a ban on so-called nudifier systems, which generate non-consensual sexually explicit images, and a prohibition on AI generating non-consensual sexual content or child sexual abuse material. Registration obligations in the EU database were reinstated for systems otherwise exempt. Support measures for SMEs were extended to small mid-cap enterprises, and the Council clarified the AI Office’s competences for supervising general-purpose AI models.
Simplification is not deregulation
It is worth being precise about what is being simplified. The substantive obligations survive; what changes is when they bite, and how much administrative machinery sits around them. Read against the record, though, a later date is only worth something if the institutions are ready when it arrives. The Act entered into force on 1 August 2024. The deadline for member states to designate their notifying and market surveillance authorities was 2 August 2025, and many missed it.
Civil society read the package the same way. A joint Equinet and ENNHRI statement on the Digital Omnibus appeared in March, alongside an open letter from TUV-Verband, AlgorithmWatch and ForHumanity.
Drawn from The EU AI Act Newsletter #98, 24 March 2026.