EU AI Act deadlines after the Digital Omnibus: a dated tracker

The EU AI Act's deadlines moved in July 2026, and one of them lands in five weeks. This tracker lists every date as the amended law now stands, what applies today, and what is still ahead. The amending law is Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, published in the Official Journal on July 24, 2026 and in force since July 27, 2026.

The dates, in order

The table below uses law firm summaries of the Regulation, because we could not retrieve the Official Journal text this run. Check the Official Journal before relying on any date for compliance. Each row names its source so you can follow it.

• August 2, 2025: Obligations for general-purpose AI models began, and per the Cooley summary they are unchanged by the Omnibus.
• August 2, 2026: Transparency obligations apply, according to Hunton. The AI Office's enforcement powers and governance body obligations also apply from this date and are unchanged, per Cooley.
• December 2, 2026: Two things land. The new Article 5 prohibitions take effect, and systems already on the market before August 2, 2026 must have machine-readable marking of AI-generated content under Article 50(2). Cooley describes the marking date as a four-month transitional period.
• August 2, 2027: National AI regulatory sandboxes, moved from August 2, 2026, per Cooley. Commission guidelines on sectoral complementarity are expected the same month, per Ecija.
• December 2, 2027: Stand-alone high-risk systems in Annex III, such as employment and education uses, moved from August 2, 2026.
• August 2, 2028: High-risk AI embedded in regulated products under Annex I, such as medical devices and machinery, moved from August 2, 2027.
• August 2, 2030: High-risk systems used by public authorities, moved from August 2, 2026, per Cooley.

The new bans

The Omnibus adds two prohibitions to Article 5. One covers AI systems that generate or manipulate intimate images, video or audio of an identifiable person without explicit consent, often called nudifier tools. The other covers systems that generate child sexual abuse material. Both take effect December 2, 2026.

Cooley's summary says that for providers, the ban applies where such output is the intended purpose, or where it is a reasonably foreseeable and reproducible outcome and the system lacks adequate safeguards. For deployers, it applies only where they use the system to generate or manipulate such material. It adds that there is "no exemption for systems already on the market."

That last point is the one to watch. A general-purpose image or video generator already on the market that can produce non-consensual intimate imagery without adequate safeguards has until December 2, 2026 to fix it or exit the EU. The wording leans on "adequate safeguards," which regulators will have to define in practice.

What changed for frontier labs

For the labs this site follows, the most relevant change is on the AI Office. Cooley reports it gets exclusive competence over AI systems built on general-purpose models where the system and the model come from the same provider or corporate group, and Hunton says the Omnibus "broadens the AI Office's supervisory role in relation to general-purpose AI models." Ecija adds that the AI Office holds exclusive powers over general-purpose models and very large platforms or search engines, including fines of up to €35 million or 7% of global turnover.

The general-purpose model rules themselves, in force since August 2025, did not move. The Omnibus is, in Cooley's words, "targeted in scope; it does not overhaul the AI Act's overall structure or risk-based approach."

What did not slip

It is easy to read "the EU delayed the AI Act" and conclude that nothing bites until 2027. That is wrong in three ways. The transparency duties that applied August 2, 2026 still apply. The general-purpose model obligations are live. And the December 2, 2026 prohibitions and marking duty are closer than anything in the high-risk package. The delays buy time for high-risk systems in hiring, education and products, which are the parts that needed standards that were not ready. They do not touch the frontier-model rules.

How this connects to the rest of the regulatory picture

The EU rules sit alongside US voluntary commitments and state laws. For a map of which rules actually bind frontier labs, see our guide to AI safety regulation. In the UK, the Online Safety Act is the tool being used against Grok deepfakes, and Ofcom's chief executive told lawmakers on October 20, 2026 that the regulator weighed an urgent court order and found the legal threshold too high, per MLex. The EU's route is different: a flat ban on the capability, with a date.

The hype check

The popular summary is that Brussels gave up on regulating AI. The dates say otherwise. The Omnibus moved some of the heaviest compliance burdens by one to four years, kept the general-purpose model regime and the transparency rules intact, and added two bans. That is a narrowing and a delay, not a retreat from the Act. Two cautions apply. The summaries we read disagree on small details, such as whether the headline name is a Regulation or an Omnibus Directive, and none of them substitutes for the text. And the next real test is enforcement: nothing in the sources we opened reports a fine under the Act yet.

What we will update

This page is a living tracker. We will update it when the Commission publishes guidance on the new prohibitions, when the AI Office takes its first enforcement step on a general-purpose model, and on December 2, 2026, when the bans and the marking duty take effect. Our position: if you ship a generative image or video product in the EU, treat December 2 as a hard date and test your safeguards against it now.