Article 111 gives transitional treatment to AI systems and general-purpose AI models that were already placed on the market or put into service before the relevant application date. In outline, high-risk systems already on the market are generally caught by the high-risk requirements only if they are subject to significant changes in design after the application date, with a separate compliance deadline for systems intended to be used by public authorities. Providers of general-purpose AI models placed on the market before 2 August 2025 must take the necessary steps to comply by 2 August 2027, and a separate Article 50(2) transition ends on 2 December 2026. Each transition is conditional, so the market date, the type and model and any later change must all be recorded.
What is a transition in the AI Act?
A transition in the AI Act is a provision that delays or limits how a new requirement applies to something that already exists. It is not a general exemption and it is not a grace period after an application date has passed. Article 111 is the main transitional provision, and it works by reference to when a system or model was placed on the market or put into service, what kind of system it is, and whether its design changes afterwards. Other transitions sit in specific provisions, such as the Article 50(2) marking transition, and each has its own conditions and end date. A team that relies on a transition should therefore be able to name the paragraph relied on, show the date and version it applies to, and explain why the conditions are met. Without that record, a transition claim is only an assumption, and it will not survive a change in the system or a question from an authority.
How does Article 111 treat high-risk systems already on the market?
Under Article 111(2), the Regulation applies to operators of high-risk AI systems, other than the large-scale IT systems covered by Article 111(1), that were placed on the market or put into service before the Chapter III application date, only if those systems are subject to significant changes in their designs from that date. That is a narrow and fact-sensitive test: the question is whether the design has significantly changed, not whether the system is still in use or has received routine maintenance. The consolidated text should be checked for how the provision treats a type and model, since the relevant date can turn on when the first unit of that type and model was placed on the market. Separately, providers and deployers of high-risk systems intended to be used by public authorities must take the necessary steps to comply by 2 August 2030, even where the general transition would otherwise apply. That public-authority position should be planned as a fixed deadline, not as an open-ended transition.
For the broader regulatory framework, see EU AI Act timeline.
What about large-scale IT systems and public authorities?
Article 111(1) deals separately with AI systems that are components of the large-scale IT systems established by the legal acts listed in Annex X, in the area of freedom, security and justice. Systems of that kind placed on the market or put into service before 2 August 2027 must be brought into compliance by 31 December 2030, without prejudice to the application of the Article 5 prohibitions, which apply without that delay. The requirements must also be taken into account when each large-scale system is evaluated under its own legal act and where those acts are replaced or amended. This is a specialised transition for named Union systems and the authorities operating them, and it should not be borrowed as a general deadline for commercial systems. A public-authority deployer outside Annex X should work from the Article 111(2) public-authority deadline of 2 August 2030 and from the deployer duties that apply to its own use.
How does Article 111 treat general-purpose AI models already on the market?
Article 111(3) provides that providers of general-purpose AI models placed on the market before 2 August 2025 must take the necessary steps to comply with the obligations in the Regulation by 2 August 2027. That date is stated in the text of Article 111 itself, so it can be planned as a fixed transition end rather than as an expected date, while the steps needed to comply still depend on the model and on whether it presents systemic risk. A provider relying on this transition should record the model's market date, version and provider identity, the Chapter V obligations assessed, and the work completed towards compliance, so the position can be shown at any point during the transition period. A downstream provider or deployer should not assume the model provider's transition extends its own duties: the downstream system has its own classification and its own application dates, and those should be recorded separately.
How does the Article 50(2) transition fit in?
The Article 50(2) transition is separate from Article 111 and narrower in scope. It gives certain providers of AI systems generating synthetic content, including some systems based on general-purpose AI models, that were already placed on the market before 2 August 2026 until 2 December 2026 to comply with the Article 50(2) marking and detection obligation. The transition concerns the provider obligation to mark outputs in a machine-readable format and make them detectable as artificially generated or manipulated, subject to the provision's conditions and exceptions. It does not postpone Article 50 generally, which began applying on 2 August 2026, and it does not postpone prohibition, literacy or general-purpose AI duties. A provider relying on it should identify the system version, the market date and the technical implementation work, and should complete verification and testing before the transition ends rather than treating the end date as the start of the work.
Continue with what Regulation (EU) 2026/1744 changed.
What does this guide not decide?
This guide does not decide whether any particular system benefits from a transition, whether a change counts as a significant change in design, or whether a system falls within Annex X or the public-authority deadline. Those questions depend on the system's history, its type and model, the nature of any change and the authority using it, and they require the consolidated text of Article 111 and, where relevant, the Annex X legal acts to be checked for the specific case. The guide also does not cover transitions in sector legislation, which can impose their own dates and conditions alongside the AI Act. Where a transition is relied on for a release, a contract or a conformity decision, confirm the position with a competent authority or legal adviser and keep the market date, version history and change record with the decision.
Frequently asked questions
Do AI systems already on the market escape the AI Act?
No. Article 111 gives conditional transitional treatment, not a general exemption. High-risk systems already on the market are generally caught only if significantly changed in design, while public-authority systems and large-scale IT system components have their own compliance deadlines, and prohibitions apply without the transition delay.
When must providers of older general-purpose AI models comply?
Under Article 111(3), providers of general-purpose AI models placed on the market before 2 August 2025 must take the necessary steps to comply by 2 August 2027.
Does the Article 50(2) transition postpone all transparency duties?
No. It concerns a limited marking and detection obligation for certain systems already placed on the market before 2 August 2026, ending on 2 December 2026. Article 50 generally began applying on 2 August 2026.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689, as amended ↗AI Act Service Desk implementation timeline ↗