EU AI Act: what applies from 2 December, and what was pushed to 2027

The next deadline of the EU's AI Act falls on 2 December 2026. From that day, a new ban on "nudifier" apps applies, and providers of older generative AI systems must mark AI-generated content. At the same time, the strictest rules for high-risk AI have been postponed by more than a year.

The changes come from the so-called Digital Omnibus on AI, which was published in the EU's Official Journal on 24 July 2026 as Regulation (EU) 2026/1744 and entered into force on 27 July.

What applies from 2 December 2026

• Ban on nudifier apps: AI systems that generate non-consensual intimate images or child sexual abuse material are prohibited. According to the European Parliament, companies have until 2 December 2026 to bring their systems into line. Legal analyses describe a safe harbour for systems with effective safeguards against such use.
• Marking AI-generated content: under Article 50(2), providers must make AI-generated audio, images, video and text detectable in a machine-readable way, for example with watermarks or metadata. For systems placed on the market before 2 August 2026, the deadline is 2 December 2026.

What was postponed

• High-risk AI under Annex III, for example systems used in hiring, credit scoring or education: obligations now start on 2 December 2027 instead of 2 August 2026.
• AI embedded in regulated products under Annex I, such as medical devices or machinery: 2 August 2028.

What did not move

The existing bans on practices such as social scoring, the AI literacy duty for companies that use AI, and the obligations for general-purpose AI models keep their original dates. The delay is a postponement of enforcement, not a removal of the requirements.

What companies should do now

• Check generative features. Any product that creates images, audio, video or text for users needs a plan for machine-readable marking.
• Check for misuse risk. Image tools need safeguards that prevent sexualised images of real people without consent.
• Use the extra year. Teams building HR, credit or education tools should use the time until December 2027 for risk management, documentation and human oversight, rather than pausing.

Canadian companies are affected too if they offer AI systems to users in the EU. The AI Act applies based on where a system is used, not where the company is based. The details of watermarking are still being worked out in a code of practice, so legal teams should follow the final guidance from the EU AI Office.