"Does the AI Act already apply?" Partly. The Regulation has been in force for over two years, but its chapters apply from different dates, laid down in Article 113, the last article of the AI Act. Since 27 July 2026 it reads as amended by the Digital Omnibus on AI, Regulation (EU) 2026/1744.
In short
The AI Act has been in force since 1 August 2024 and, as a general rule, has applied since 2 August 2026 (Article 113). Articles 4 and 5 have applied since 2 February 2025, Article 50 since 2 August 2026, and the new prohibitions in Article 5 apply from 2 December 2026. The high-risk rules apply from 2 December 2027 (Annex III) and 2 August 2028 (Annex I), dates set by the Digital Omnibus. Article 111 covers existing systems. The GDPR applies separately, since 25 May 2018.
Who the AI Act applies to is covered in the scope of Article 2; this piece is about when.
Entry into force is not the same as application
The first paragraph of Article 113 governs entry into force: the twentieth day following publication. Publication took place on 12 July 2024 (OJ L, 2024/1689), so the AI Act has been in force since 1 August 2024. But an obligation arises only once the article containing it also applies, which the second and third paragraphs govern.
Regulation (EU) 2026/1744 was published on 24 July 2026 and entered into force on 27 July 2026 (Article 4 of Regulation (EU) 2026/1744). In our reading, an amended article takes effect only once the article itself applies: the new Article 4 since 27 July 2026, because Chapter I already applied; the amended high-risk rules from the dates in point (c).
The dates of application in one table
| Date | What applies | Legal basis | Mainly for |
|---|---|---|---|
| 1 August 2024 | Entry into force of the Regulation; no obligations for organisations yet | Article 113, first paragraph | Everyone: the clock starts |
| 2 February 2025 | Chapter I (including Article 4, AI literacy) and Chapter II (Article 5, prohibited practices) | Third paragraph, point (a) | Providers and deployers of all AI systems |
| 2 August 2025 | Chapter III Section 4 (notifying authorities and notified bodies), Chapter V (general-purpose AI models), Chapter VII (governance), Chapter XII (penalties, except Article 101) and Article 78 (confidentiality) | Third paragraph, point (b) | Model providers, Member States, conformity assessment bodies |
| 27 July 2026 | Articles 102 to 110 (amendments to other Union acts) | Third paragraph, point (d) (new) | Mainly the Commission when adopting acts for sectors such as aviation, vehicles and rail; Article 110 adds the AI Act to the Annex to Directive (EU) 2020/1828 on representative actions |
| 2 August 2026 | General rule: everything the third paragraph does not mention, including Chapter IV (Article 50, transparency), Chapter VI (innovation), Chapter IX (monitoring and market surveillance), Article 101 and Article 6(5) | Second paragraph | Providers and deployers of chatbots, generative AI and deep fakes; the Commission towards model providers |
| 2 December 2026 | New prohibitions: AI that generates or manipulates realistic intimate or sexually explicit images, videos or audio of an identifiable person without explicit consent, and AI that generates child sexual abuse material (Article 5(1), first subparagraph, points (ba) and (bb), with paragraphs (1a) and (1b)) | Third paragraph, point (a) (new) | Providers and deployers of image, video and audio generators; for providers only where such output is the intended purpose or foreseeable without adequate safeguards (paragraph (1a)) |
| 2 December 2027 | Chapter III, Sections 1, 2 and 3 (except Article 6(5)) for systems classified as high-risk pursuant to Article 6(2) and Annex III | Third paragraph, point (c)(i) | Among others public authorities, banks, insurers, employers and education institutions with Annex III systems |
| 2 August 2028 | Chapter III, Sections 1, 2 and 3 (except Article 6(5)) for systems classified as high-risk pursuant to Article 6(1) and Annex I | Third paragraph, point (c)(ii) | Manufacturers and providers of regulated products, such as medical devices and toys |
Three things stand out. Article 50 is not among the exceptions and has therefore applied since 2 August 2026; the Omnibus did not change that date, but it did give existing generative systems until 2 December 2026 for the marking (Article 111(4)). Article 101, the Commission's fines for model providers, is excluded from point (b) and has applied since 2 August 2026. And Article 6(5), the task for the Commission to provide guidelines on classification, is outside point (c) and in our reading follows the general rule (see the missed deadline for those guidelines).
What the Digital Omnibus changed in Article 113
Article 1, point 40, of Regulation (EU) 2026/1744 amends only the third paragraph of Article 113; 1 August 2024 and 2 August 2026 remain unchanged. Three amendments:
- Point (a) is replaced. Chapters I and II still apply from 2 February 2025, but the prohibitions the Omnibus adds to Article 5 (points (ba) and (bb), paragraphs (1a) and (1b)) apply from 2 December 2026.
- Point (c) is replaced. The original text set a later date only for Article 6(1) and the corresponding obligations: 2 August 2027. The rest of Chapter III, including Annex III, fell under the general date of 2 August 2026; only Section 4 already applied from 2 August 2025 (point (b)). Now Sections 1, 2 and 3 apply from 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
- Point (d) is added: Articles 102 to 110 apply from 27 July 2026, the day the Omnibus entered into force (see the AI Act for cars, aviation, rail and ships).
Article 1, point 39, of Regulation (EU) 2026/1744 aligned Article 111(2) with this: the reference date for existing high-risk systems was 2 August 2026 and is now the date of application of Chapter III. The same point added Article 111(4).
Recital 40 of Regulation (EU) 2026/1744 gives the reason for point (c): the delayed availability of standards, common specifications and alternative guidance, and the delayed establishment of national competent authorities. The dates are fixed and do not depend on the standards. More in high-risk AI postponed to 2 December 2027.
Dates that are not in Article 113
Some articles carry their own deadline, such as Article 57(1): at least one AI regulatory sandbox per Member State operational by 2 August 2027.
Registration (Articles 49 and 71). Neither Article 49 (Chapter III, Section 5) nor Article 71, the EU database (Chapter VIII), is mentioned in the third paragraph of Article 113. According to the text, Article 49 has therefore applied since 2 August 2026. In our reading, the obligation only has an object from 2 December 2027: registration concerns Annex III systems, and that classification applies from that date. An exception is 'real-time' remote biometric identification by law enforcement: there, registration under Article 49 is already a condition (Article 5(2)). See registering AI systems in the EU database.
Existing systems: Article 111
Article 111 sets out what the dates mean for systems that already existed. The main line: a high-risk AI system placed on the market or put into service before the date of application of Chapter III becomes subject to the high-risk rules only if, as from that date, it is subject to significant changes in its design (paragraph 2; the text says "this Regulation"). Recital 39 of Regulation (EU) 2026/1744 says this works per type and model: if at least one unit was lawfully placed on the market or put into service before that date, the rule also covers later units whose design remains unchanged. Systems intended to be used by public authorities must in any case comply by 2 August 2030. Article 5 always applies. General-purpose AI models placed on the market before 2 August 2025 have until 2 August 2027 (paragraph 3). The details are in the transitional rules of Article 111.
Example 1: the municipality
A mid-sized municipality has: a supplier's chatbot on its website, a tool the communications team uses to draft news items, an AI assistant for staff, a system that has been pre-sorting applications for special assistance since 2023, and a new system for a low-income scheme due to go live in 2028. The example is fictitious.
What already applies.
- AI literacy. Since 2 February 2025, and in this wording since 27 July 2026, the municipality as deployer takes measures to support the AI literacy of its staff and other persons dealing with the operation and use of AI systems on its behalf (Article 4). It does not have to guarantee any specific level for each individual. See the evidence benchmark for Article 4.
- Prohibited practices. Since 2 February 2025, no system may fall under Article 5, for example as social scoring (paragraph 1, point (c)), the 2023 system included.
- Transparency. The duty to inform people that they are interacting with an AI system rests on the provider of the chatbot and has applied since 2 August 2026 (Article 50(1)). If the municipality builds the chatbot itself and puts it into service under its own name, it is both provider and deployer (Article 3, points 3 and 4). AI-generated text published to inform the public on matters of public interest must be disclosed, unless it underwent human review or editorial control and someone holds editorial responsibility (Article 50(4)). See disclosing a chatbot as AI.
- The GDPR. Where these systems process personal data, the GDPR has applied since 25 May 2018 (Article 99(2) GDPR), whatever date the AI Act sets. Processing likely to result in a high risk requires a data protection impact assessment (Article 35 GDPR).
What comes from 2 December 2027. The new system for the low-income scheme evaluates whether residents are eligible for public assistance benefits and is therefore high-risk in principle (Annex III, point 5(a)), unless Article 6(3) provides an exception. If it is put into service after 2 December 2027, the deployer obligations (Article 26), the prior fundamental rights impact assessment (Article 27(1)) and the registration of its use (Article 49(3) and Article 26(8)) apply. That does not hold if the same type and model was already placed on the market or put into service before 2 December 2027: this system then also falls under Article 111(2) (recital 39 of Regulation (EU) 2026/1744) and, as a system intended to be used by a public authority, must comply by 2 August 2030. If the 2023 system is also high-risk, it is different: under Article 111(2), the high-risk obligations, including the fundamental rights impact assessment, apply only if its design is significantly changed as from 2 December 2027, and because it is intended to be used by a public authority, by 2 August 2030 at the latest.
Which Dutch authority will supervise the AI Act has not been settled: the Uitvoeringswet AI-verordening (the Dutch implementing act) is a draft, with submission to parliament planned for the second quarter of 2027 (Kamerstuk 26643, no. 1554).
Example 2: the bank and the insurer
Since 2024 a bank has used a creditworthiness model for private customers; it has built its own customer chatbot and is considering software that measures the emotions of call centre staff. Also fictitious.
- Now. The chatbot is the bank's own: it is the provider and, since 2 August 2026, informs customers that they are interacting with AI (Article 50(1)). The emotion software in the workplace, where it infers emotions from biometric data such as voice or face, has been prohibited since 2 February 2025, except for medical or safety reasons (Article 5(1), point (f), and Article 3, point 39). Article 4 applies here too.
- From 2 December 2027. Evaluating the creditworthiness of natural persons is high-risk in principle (Annex III, point 5(b); detecting financial fraud is excluded). For a new model, the deployer obligations (Article 26) and the fundamental rights impact assessment of Article 27 then apply; if the bank develops the model itself and puts it into service under its own name, it is also the provider, and the requirements of Section 2 and the provider obligations of Section 3 apply as well. For the 2024 model, Article 111(2) applies: only on a significant change in its design. The 2 August 2030 deadline applies only to systems intended to be used by public authorities; in our reading a bank does not fall under that, so a model from before 2 December 2027 without significant changes in its design does not become subject to the high-risk obligations.
An insurer with a pricing model for life and health insurance (Annex III, point 5(c)) is on the same track.
What to do now
- Record two dates per system in your AI register: when it (its type and model) was placed on the market or put into service, and which date in Article 113 applies.
- Complete what already applies: Article 4, Article 5 and Article 50, and the GDPR.
- Mind 2 December 2026 if you provide or deploy systems that generate images, videos or audio: the new prohibitions in Article 5 apply, and generative systems placed on the market before 2 August 2026 must mark their output in a machine-readable format (Article 50(2) and Article 111(4)).
- Plan the high-risk work towards 2 December 2027 or 2 August 2028, and record what a significant change in design means for you.
The case check shows which AI Act and GDPR obligations come into view for a use case.
Frequently asked questions about the AI Act's dates of application
Sources
Newsletter
Every Tuesday, the AI Act week ahead in 5 minutes
A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.
Practical and short · No spam · One-click unsubscribe