Deadlines and primary sources for the EU AI Act transparency rules, maintained by the team behind the scanner. Every entry is read from the official texts. When the law moves, this page logs it.
Entries last verified against the primary sources on 2026-09-13
How this page is written: AI drafts the summaries, and a person checks every date, quote and link against the primary source before it goes live. Nothing here is auto-published, including the correction log below.
Next deadline
Marking grace period ends for pre-existing generative systems
The regulation is law from this date; obligations switch on in stages listed below.
Source: Regulation (EU) 2024/1689, Art. 113
Banned AI practices (Art. 5) and the duty to ensure staff AI literacy (Art. 4) apply to providers and deployers — including businesses that merely use AI tools.
Source: Art. 113(a)
18:00 CEST cut-off to appear on the list of initial signatories published before 2 August 2026; about 190 organisations made it. Signing itself did NOT close — providers and deployers may sign at any time, including after this date, and the published list keeps being updated.
Source: European Commission, digital-strategy — signing FAQ
Visitors must be told when they interact with an AI system (e.g. a chatbot); deep fakes and AI-written text published on matters of public interest must be disclosed as such. The Digital Omnibus did NOT postpone this date. Fines: up to €15 m / 3 % of turnover; SMEs pay the lower of the two.
Source: Art. 113; Art. 99(4), 99(6)
Providers of generative AI systems placed on the market before 2 August 2026 must meet the Art. 50(2) machine-readable marking duty by this date.
Source: Art. 111(4), inserted by Regulation (EU) 2026/1744 (Digital Omnibus on AI)
Postponed by the Digital Omnibus from the original 2026 date. Relevant if your product itself is a high-risk AI system — not for ordinary websites.
Source: Art. 113(3)(c), as amended
AI safety components in regulated products (machinery, medical devices, etc.) — the last major application date.
Source: Art. 113, as amended
In motion: tracked, no fixed date yet
8 documents · verified 2026-09-13
Commentary ages; the texts do not. These are the documents this timeline and the scanner's checks are built on.
The law itself, in all EU languages. It is a regulation — directly applicable in every member state, no national transposition needed for it to bind you.
The article behind this scanner: people must be told when they interact with an AI system, and certain AI-generated content (deep fakes, and AI-written text published on matters of public interest) must be disclosed as such. Annotated per-paragraph view.
Article 50 breaches: fines up to €15 million or 3 % of worldwide annual turnover, whichever is higher — for SMEs, whichever is lower.
Where guidelines, codes of practice and official timelines land first. The pages this watch monitors for changes.
The final Article 50 guidelines, published 20 July 2026 — the Commission's own reading of what counts as adequate disclosure. Non-binding, but it is what a market-surveillance authority reaches for first.
Voluntary code (final version 10 June 2026) describing how signatories mark and disclose AI-generated content in practice. Sections may be signed separately — a deployer that only publishes AI-generated content signs Section 2 — and the code stays open for signature.
The amending regulation of 8 July 2026, in the Official Journal since 24 July 2026. It is the legal basis for the postponed high-risk dates and the marking grace period below — Article 50's own date it left untouched.
Enforcement is national. Example from our home market: Lithuania's communications regulator RRT supervises AI Act compliance and serves as the single contact point.
It binds providers and deployers of AI systems in scope, for example a chatbot that talks to visitors (Art. 50(1), a provider duty) or AI-generated text published to inform the public on matters of public interest (Art. 50(4), a deployer duty). Whether you are a provider or a deployer of your particular setup is a judgement for you and your advisers; this page gives you the dates and the primary texts to make it with.
The Article 50 transparency obligations started to apply. The Digital Omnibus (Regulation (EU) 2026/1744, in the Official Journal since 24 July 2026) postponed several high-risk dates but left Article 50's date untouched. Penalties reach €15 million or 3 % of worldwide annual turnover, whichever is higher; for SMEs, whichever is lower.
It is voluntary and it remains open: providers and deployers may sign at any time, including after 27 July 2026. That date closed only the list of initial signatories published before 2 August 2026. Sections can be signed individually: a deployer that publishes AI-generated content signs Section 2.
No. The guidelines published on 20 July 2026 are non-binding, but they set out how the Commission reads Article 50, which is what a market-surveillance authority reaches for first. The Commission and the AI Board also concluded, on 8 and 9 July 2026, that the Code of Practice is an adequate tool for implementing Article 50(2), (4) and (5).
Scope of this page
An information service: dates and documents summarised from official sources, each entry linked so you can verify it yourself. It is not legal advice, and a summary never overrides the linked text. What Article 50 means for your specific setup is a judgement for you and your advisers.