Brussels: The AI Act omnibus reaches its first hard deadline on 2 December 2026, when the ban on nudification apps and the watermarking duty for older generative systems both start to apply. Parliament approved the deal on 16 June 2026 and the Council gave its final green light on 29 June, according to compliance specialist Usercentrics. Providers now have about three and a half weeks to prepare.
Negotiators struck a provisional agreement in early May 2026 after a first trilogue on 28 April failed, Gibson Dunn reports in its client alert. Member state representatives confirmed the compromise on 13 May. The AI Act omnibus rewrote the timetable for the world’s first horizontal AI law.
Stand-alone high-risk systems listed in Annex III now fall under the rules from 2 December 2027, while AI embedded in regulated products under Annex I follows on 2 August 2028. Both dates sit later than the original schedule. Member states must also have AI regulatory sandboxes in place by 2 August 2027.
Transparency duties did not move. Under Article 50, disclosure that a person is talking to an AI system and the labelling of deepfakes have applied since 2 August 2026, and Gibson Dunn notes that these obligations “remain unaffected and proceed as scheduled from 2 August 2026.” Article 50(2) gives providers of systems already on the market before that date a grace period for machine-readable watermarking until 2 December 2026. Systems placed on the market on or after 2 August get no grace period, Usercentrics adds.
The deal also adds a new prohibition in Article 5 on AI systems that generate non-consensual intimate imagery and child sexual abuse material. The AI Act omnibus ban reaches so-called nudifier apps where such output is a reasonably foreseeable and reproducible result without significant technical modification. Gibson Dunn says the transitional period runs until 2 December 2026, and Usercentrics describes a safe harbour for systems whose technical safeguards reliably prevent such outputs.
For companies, the practical task is to sort systems by date. Providers of older generative models covered by the AI Act omnibus must add watermarking before 2 December 2026, and developers of image tools should test how easily their products can produce prohibited content. Teams that planned around the earlier high-risk dates gain time, but the AI Act omnibus does not excuse them from documentation and risk work.
Readers should check dates against the primary texts. Law firm and vendor summaries differ on the exact day of the provisional deal, citing both 6 May and 7 May 2026, and the Council’s 29 June press release on final approval is the reference for the adoption step. Gibson Dunn’s analysis of the postponed deadlines sets out the compromise in detail.
The next milestone is 2 December 2026. By then the nudification ban and the watermarking duty will bite, and the AI Act omnibus will have moved from negotiation to enforcement work for national authorities and providers alike.





