NO AI ACT.

AI Act · Timeline · Official sources

The AI Act application timeline

"Is the AI Act already in force?" It is the most common classroom question — and the answer is: yes, but its rules apply in stages. Here are the dates that matter, what already applies and what comes next, in plain language and with official sources.

01The stages at a glance

The AI Act does not apply all at once: it entered into force on 1 August 2024 and its rules become applicable in stages (Art. 113). The key dates:

Staged application of Regulation (EU) 2024/1689
DateWhat becomes applicable
1 August 2024Entry into force of the regulation (not yet the application of its rules).
2 February 2025General provisions, the prohibited practices as they then stood, and the AI-literacy obligation (Art. 4). The two practices added in 2026 follow their own date, below.
2 August 2025Rules on GPAI models, governance, notifying authorities and penalties.
2 August 2026General application, including the Article 50 transparency duties — with one exception: the synthetic-content marking duty of Article 50(2), which providers of generative systems already on the market before this date have until 2 December 2026 to meet (Art. 111(4)). The rest of Article 50, including the deepfake disclosure, applies to everyone from this date. Not the requirements for high-risk systems, deferred by the Digital Omnibus (see below).
2 December 2026Two new prohibited practices added to Article 5 by the Digital Omnibus. Same date for the synthetic-content marking duty on generative systems already on the market before August 2026.
2 August 2027Compliance deadline for GPAI models placed on the market before August 2025. By the same date every Member State must have an operational regulatory sandbox.
2 December 2027Requirements and obligations for Annex III high-risk systems (Art. 6(2)): employment, education, essential services, justice.
2 August 2028Requirements and obligations for high-risk systems embedded in already-regulated products (Annex I, Art. 6(1)).
2 August 2030Deadline for high-risk systems intended for public authorities already on the market.

02Entry into force ≠ application

These are two different moments: entry into force (1 August 2024) makes the regulation part of EU law; application is when obligations actually bind those who build or deploy AI systems. The staged calendar gives time to adapt — the most urgent rules first (the prohibitions), then the bulk of the obligations.

The penalty calendar does not coincide with the obligation calendar: the difference is set out on AI Act penalties.

03What already applies (as of the review date)

As of this page's last review (September 2026) the following apply: the bans on unacceptable practices, the AI-literacy obligation, the GPAI rules and, since 2 August 2026, general application together with the Article 50 transparency duties.

The requirements for high-risk systems, by contrast, are not yet enforceable. Regulation (EU) 2026/1744 of 8 July 2026 — the "Digital Omnibus" — deferred them to 2 December 2027 for Annex III and to 2 August 2028 for systems embedded in already-regulated products (Annex I). The stated reason is the unavailability of standards and guidance and the delay in designating national competent authorities.

Deferred does not mean repealed. The obligations are unchanged in substance: what moved is the date from which they can be demanded. Anyone designing a high-risk system today has more time to comply, not fewer duties.

The same regulation adds two prohibited practices to Article 5, applicable from 2 December 2026: systems that generate or manipulate realistic intimate imagery of an identifiable person without their explicit consent, and systems that generate child sexual abuse material. On this front the framework grew wider, not lighter. Both bans carry precise scope limits, however — the system's purpose, the foreseeability of the abuse, the absence of safeguards, the deployer's own purpose — set out on the prohibited practices page.

The dates are fixed by the regulation itself; practical implementation (national authorities, guidance, standards) is still evolving — check the official sources below for the current state.

On what to tackle first inside a company, see AI Act for business.

04Where the game comes in

NO AI ACT flips this calendar on its head: its 2032 city is a world where none of these dates ever arrived. Playing a case — for instance the scoring city — makes concrete what the 2025 bans and the 2026 obligations are trying to prevent.

Play a case EU AI Act guide

05Official sources and transparency