Regulation (EU) 2024/1689 applies in phases. Regulation (EU) 2026/1744, in force since 27 July 2026, moved some of those dates. This page shows the resulting timeline, indicating which instrument set each milestone, together with the articles and who they are addressed to.
The marked dates come from Regulation (EU) 2026/1744 (Digital Omnibus on AI), published in the Official Journal of the European Union on 24 July 2026 and in force since 27 July 2026, which amends Regulation (EU) 2024/1689.
- Regulation in force (Reg. (EU) 2024/1689)
- Dates amended by Regulation (EU) 2026/1744
The 8 key dates of the EU AI Act, one by one
- In force
Art.4 · Art.5
2 February 2025 · AI literacy and prohibited practices
AI literacy (Art. 4): providers and deployers take measures to support the development of AI literacy of their staff and of others operating AI systems on their behalf (wording as given by Regulation (EU) 2026/1744, in force since 27 July 2026; the obligation applies from 2 February 2025). Prohibited practices (Art. 5): practices such as subliminal techniques, social scoring, inferring emotions in the workplace or in education, or the untargeted scraping of facial images are kept off the market. Both articles are addressed to providers and deployers alike.
Related terms: AI literacy · AI system · Deployer · Minimal risk · Prohibited practices · Provider
- In force
Chapter V · Chapter VII · Chapter XII (in part) · Art.78
2 August 2025 · General-purpose AI (GPAI) models and European governance
Obligations of providers of general-purpose AI models — technical documentation, copyright policy, summary of the training content (Chapter V). The European governance structure is set up (AI Office, scientific panel) and the penalties regime enters into application in part. The GPAI obligations in Chapter V fall on the provider of the model; whoever uses it as a deployer does not take them on.
Related terms: GPAI — General-purpose AI models · Penalties regime (Article 99)
- In force
Art.50(1) · Art.50(3) · Art.50(4)
2 August 2026 · Transparency: chatbots, emotion recognition and generated content
AI systems that interact directly with natural persons are identified as such (50.1, provider). Permitted emotion recognition or biometric categorisation systems inform the persons exposed to them (50.3, deployer). Deep fakes and AI-generated text on matters of public interest are labelled (50.4, deployer). This date stands under the Regulation in force, unchanged by the Omnibus.
Related terms: AI system · Deployer · Marking of synthetic content · Provider · Transparency obligations (Article 50)
- LCReg. (EU) 2026/1744
Art.50(2) · Art.5.1 (ba) and (bb)
2 December 2026 · Watermarking and two new prohibited practices
Machine-readable marking of synthetic content (50.2, provider): Art. 111(4) provides a transitional period until this date only for systems already placed on the market before 2 August 2026 — new systems have no transitional period. Two further prohibited practices are also added to Art. 5.1, points (ba) and (bb): the generation or manipulation of non-consensual intimate material and of child sexual abuse material. For the deployer the prohibition operates only where it uses the system for that purpose.
Related terms: Prohibited practices
- LCReg. (EU) 2026/1744
Art.111(3) · Art.57(1)
2 August 2027 · GPAI transitional period and national sandboxes in operation
Two distinct matters fall on this date, and neither is an obligation of the deployer. Art. 111(3) transitional period: general-purpose AI models placed on the market before 2 August 2025 must take the necessary steps to comply with Chapter V by this date at the latest — it binds the provider of the model, not whoever uses it; for the deployer it is a matter of documentary verification of its supplier. Regulatory sandboxes (Art. 57(1)): Member States must have at least one AI regulatory sandbox operational at national level by this date; the original timeline set 2 August 2026 and Regulation (EU) 2026/1744 moved it here. It is a deadline addressed to the Member States: whether a sandbox will be available in a given Member State on that date, and on what access conditions, depends on its national implementation.
- LCReg. (EU) 2026/1744
Annex III · Art.26 · Art.27
2 December 2027 · High risk (Annex III): the first of Chapter III’s two dates
Chapter III, Sections 1, 2 and 3 — the requirements for high-risk AI systems and the obligations of their operators, Articles 6–7, 8–15 and 16–27 — does not have one application date: it has two, and both cover the same block of obligations. This is the first: 2 December 2027, for systems classified as high-risk under Article 6(2) and Annex III (recruitment, credit scoring, education, among other areas). It comprises the deployer obligations (Art.26) — use in accordance with the instructions for use drawn up by the provider, human oversight, keeping of logs — and the fundamental rights impact assessment (Art.27, FRIA) for certain deployers, mainly public bodies and providers of certain services. Art.26 is not confined to this date: for Annex I high-risk systems it applies on 2 August 2028, in the next entry. Art.27 is: its paragraph 1 limits its scope to the systems referred to in Article 6(2), so 2 August 2028 does not operate on it. This is set by Article 113, third paragraph, point (c), (i) and (ii), as amended by Regulation (EU) 2026/1744; point (c) excepts Article 6(5). The original timeline set these obligations at 2 August 2026; Regulation (EU) 2026/1744 moved them to this date.
Related terms: AI Officer · AI system · Deployer · DPIA — Data protection impact assessment · FRIA — Fundamental rights impact assessment · High risk · Provider · Substantial modification
- LCReg. (EU) 2026/1744
Art.6.1 · Annex I
2 August 2028 · High risk (Annex I): the same obligations, second date
The second date of the same block. Chapter III, Sections 1, 2 and 3, for systems classified as high-risk under Article 6(1) and Annex I: AI systems acting as safety components of products with their own harmonised legislation — medical devices, machinery, toys, among others. This is not a separate category of obligations: it is the same Chapter III as the previous entry, and it therefore comprises Article 26 — deployer obligations — which here applies from 2 August 2028 and not from 2 December 2027. Article 27 — the fundamental rights impact assessment — does not: its paragraph 1 only reaches the systems referred to in Article 6(2), so none of the Annex I systems falls within its scope and its only application date remains 2 December 2027. And there is a third destination that no date expresses: never. For high-risk systems under Article 6(1) relating to products covered by Annex I, Section B, Article 2(2), as amended by Regulation (EU) 2026/1744, provides that only Article 6(1), Article 60a and Articles 102 to 112 apply to them — and Articles 57, 58 and 59 only to the extent that the requirements have been integrated into those Union harmonisation acts: Article 26 and Article 27 never reach them.
Related terms: High risk · Substantial modification
- LC
Art.111(2)
2 August 2030 · High-risk systems of public authorities deployed before 2 December 2027
End of the timeline. High-risk AI systems intended to be used by public authorities that were already deployed before 2 December 2027 must take the necessary steps to comply with this Regulation by this date at the latest (Art. 111(2)). It is the only deadline that reaches pre-existing public-sector systems; pre-existing private systems are caught only if they are subject to significant changes in their design.
Which article applies to a given organisation depends on its role (provider or deployer) and on each AI system. The badges indicate the regulatory nature of each milestone (LC: legal obligation with a confirmed future date); this page describes the Regulation and is not legal advice.