Timeline
EU AI Act timeline after the Digital Omnibus: what applies when
Short answer: Article 50 applies now. Generative systems on the market before 2 August 2026 must mark their output by 2 December 2026. Annex III high-risk duties start on 2 December 2027, and Annex I duties on 2 August 2028. The GDPR already applies to any agent that handles personal data.
The Digital Omnibus, Regulation (EU) 2026/1744, has been in force since 27 July 2026. It moved the dates of the high-risk chapter of the EU AI Act, Regulation (EU) 2024/1689, gave older generative systems until 2 December 2026 to mark their output, and added two prohibitions to Article 5. This guide gives each date as the consolidated text states it, says who it binds, and lists what a company that runs AI agents should have on file.
The dates at a glance
What applies today
Chapters I and II have applied since 2 February 2025. They hold the definitions, the Article 5 prohibitions and the Article 4 duty on AI literacy. That duty is best-efforts: you take measures to build your staff's AI literacy, without guaranteeing what each person knows. The rules on general-purpose AI models, governance and penalties followed on 2 August 2025. On 2 August 2026 the rest became applicable, except what Article 113 defers, and the Commission gained its power to fine general-purpose model providers (Article 101).
A system that talks to people must say it is an AI, unless that is obvious. The Article 50 guide takes that duty paragraph by paragraph. Article 50(5) sets the moment:
The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure. (Article 50(5))
Two reference texts arrived just before that date: the Code of Practice on Transparency of AI-Generated Content (10 June 2026) and the Commission Guidelines on Article 50 (20 July 2026). The Code of Practice is voluntary. A provider that does not follow it must show another way the marking duty is met.
The full Article 113 schedule
The schedule below comes from the same date table Vidimus prints in its evidence packs. Our changelog reports any change to it month by month.
- Applicable
Chapters I–II: definitions, the Article 5 prohibited-practice list, and the Article 4 AI-literacy duty.
- Applicable
GPAI model obligations (Chapter V), governance (Chapter VII), notified bodies (Chapter III Section 4), and penalties. Legacy GPAI models placed before this date: 2 Aug 2027.
- Applicable
Article 50 transparency duties, Chapter IX market surveillance and Article 101 (Commission fines on GPAI providers).
- Pending
Article 50(2) machine-readable marking grace ends for systems placed on the market before 2 Aug 2026. The two new Article 5 prohibitions, 5(1)(ba) and 5(1)(bb), apply.
- Pending
Regulatory sandboxes (Article 57) and the legacy-GPAI deadline (models placed before 2 Aug 2025 must be compliant).
- Pending
Chapter III Sections 1, 2 and 3 for Annex III stand-alone high-risk systems: classification, the high-risk requirements (Arts 8–15), and the provider/deployer obligations (Arts 16–27).
- Pending
Chapter III for Annex I embedded (product-safety-component) high-risk systems.
- Pending
High-risk AI systems intended to be used by public authorities, placed on the market or put into service before the date of application of Chapter III (2 Dec 2027 for Annex III, 2 Aug 2028 for Annex I): compliance by 2 Aug 2030.
- Pending
AI systems that are components of the large-scale IT systems listed in Annex X (Freedom, Security and Justice).
Disputed provisions that may already apply
Three groups of provisions are disputed after the Omnibus. Each can be read as applying since 2 August 2026, but each only works once a system is high-risk, and the high-risk dates moved.
These may already apply. Sources disagree after the Omnibus, so ask counsel before relying on the later date.
The Article 73 reporting deadlines are short when they do apply: 15 days as a rule, two days for a widespread infringement and 10 days where a person has died.
What changes on 2 December 2026
Two things change, and the 2 December 2026 guide covers both. The first is a deadline for generative systems already on the market:
Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, that have been placed on the market before 2 August 2026 shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026. (Article 111(4))
Article 50(2) asks for output marked in a machine-readable format and detectable as artificial. A system placed on the market since 2 August 2026 already owes it. An assistant placed on the market before that date that drafts customer letters must carry the mark by 2 December 2026, unless it only assists with standard editing.
The second change is to Article 5. Points (ba) and (bb) of Article 5(1) ban systems that generate intimate imagery of an identifiable person without explicit consent, and systems that generate child sexual abuse material. Under Article 5(1a), placing such a system on the market is banned in two cases. Either that output is its intended purpose, or it is a foreseeable and reproducible result of a design without adequate safeguards. Using one is banned when the deployer uses it for that purpose.
Annex III high-risk duties from 2 December 2027
The consolidated Article 113 reads:
Chapter III, Sections 1, 2, and 3, with the exception of Article 6(5), shall apply from: (i) 2 December 2027 as regards AI systems classified as high-risk pursuant to Article 6(2) and Annex III; and (ii) 2 August 2028 as regards AI systems classified as high-risk pursuant to Article 6(1) and Annex I (Article 113, point (c))
Sections 1 to 3 hold the classification rules of Article 6, the requirements of Articles 8 to 15 and the duties of providers and deployers in Articles 16 to 27. For most companies, the way in is Annex III, point 4: an agent that screens candidates or evaluates staff. Read the Annex III guide, the deployer guide and the Article 14 guide on human oversight. Our high-risk page shows what a review involves.
One rule on systems already running matters for planning:
this Regulation shall apply to operators of high-risk AI systems... that have been placed on the market or put into service before the date of application of Chapter III referred to in Article 113, only if, as from that date, those systems are subject to significant changes in their designs. (Article 111(2))
An HR screening agent in production before 2 December 2027 stays outside Chapter III until its design changes significantly. A new model or a new vendor may be such a change: decide with counsel and write the reasoning down.
Annex I from 2 August 2028
This date covers AI that is a safety component of a product under the Union laws listed in Annex I, where that product needs a third-party conformity assessment (Article 6(1)). Since the Omnibus, a system used only for functions unrelated to safety, such as user assistance, automation or convenience, is not a safety component (Article 6(1a) and (1b)). The exception is a system whose failure would endanger health and safety. Most software teams never provide one, but a manufacturer that builds AI into a machine or a medical device may.
What the deferral does not change
For Annex III systems, the Omnibus moved dates, not the line between risk classes. Article 6 applies from 2 December 2027, but its criteria are already written, so a system can be classified today. Article 6(3) lets an Annex III system out of the high-risk class where it poses no significant risk of harm, with one hard limit:
Notwithstanding the first subparagraph, an AI system referred to in Annex III shall always be considered to be high-risk where the AI system performs profiling of natural persons. (Article 6(3))
An agent that pre-screens job applications, or scores the performance of staff, should be classified now, and the written assessment that Article 6(4) asks for kept on file.
No harmonised standard has been cited in the Official Journal, so there is no presumption of conformity and nobody can certify an AI system against the AI Act today. Vidimus is not a notified body. Compliance remains a decision your organisation makes and defends on evidence, and the evidence primer says what that file holds.
Evidence of behaviour comes from tests. Vidimus sends each test ten times, because an agent can answer the same question differently each time, and one reply proves nothing. The finding is a pass rate, and how we test sets out the method.
The GDPR, Regulation (EU) 2016/679, has applied since 25 May 2018. An agent that answers customers or reads their records processes personal data today, whatever its AI Act class. The GDPR guide says which of its duties a test can show.
To see which of these dates binds your agent, run the free AI Act risk classifier.
Quick answers
Does the AI Act already apply to our chatbot?
Yes. Article 50 has applied since 2 August 2026 to any AI system that talks to people, and the Article 5 prohibitions since 2 February 2025. Only the high-risk chapter was deferred.
When do the Annex III high-risk duties start?
On 2 December 2027, under Article 113 as amended by Regulation (EU) 2026/1744. Annex I products follow on 2 August 2028. A system already in service stays outside Chapter III until its design changes significantly.
Which provisions are disputed?
The disputed provisions are Articles 40 to 49, Articles 72 and 73, and Article 86. Sources disagree on whether they apply before the high-risk dates, so ask counsel before relying on the later date.
Can anyone certify our AI Act compliance today?
No. No harmonised standard has been cited in the Official Journal, so there is no presumption of conformity. Vidimus is not a notified body: it produces evidence for your own decision.
Does the GDPR apply alongside the AI Act?
Yes. The GDPR has applied since 25 May 2018 to any processing of personal data, and the AI Act does not replace it. An agent that talks to customers usually owes both.
Sources
Vidimus classifies your agent, tests the live agent with ten attempts per test, and files the result in a signed evidence pack.