What ends on 2 December 2026 for Article 50 marking?

Which systems does the transition cover?

The transition covers, in outline, providers of AI systems that generate synthetic content and that were already placed on the market before 2 August 2026, for the Article 50(2) marking and detection obligation. The key facts are the placing date and the system version: the provider must be able to show that the system was placed on the market before 2 August 2026, and must identify the version to which the transition is claimed. A system placed on the market on or after 2 August 2026 does not benefit from the transition for that obligation, and must meet the marking duty from 2 August 2026. A new version that amounts to a new placing should be assessed separately, with the placing date for that version recorded. The transition file should contain the placing evidence, the version history, and the plan to meet the duty by 2 December 2026.

Which obligation is transitioned, and which are not?

Only the Article 50(2) marking and detection obligation is transitioned to 2 December 2026 for the systems described above. Article 50(1) interaction disclosure, Article 50(3) deployer duties for emotion recognition and biometric categorisation, and Article 50(4) deepfake and public-interest text duties apply from 2 August 2026, without the same transition. A provider that treats 2 December 2026 as the date for all transparency work will have missed the earlier duties. The compliance plan should therefore list each Article 50 paragraph separately, with its application date, its status, and its owner, and should show that the earlier duties were met from 2 August 2026 while the marking work was completed by 2 December 2026. The plan should be dated and reviewed, so that the transition is shown as a managed timetable rather than a delay.

For the broader regulatory framework, see Transparency duties for AI.

What must be ready by 2 December 2026?

By 2 December 2026, the transitional systems must meet the Article 50(2) obligation: outputs must be marked in a machine-readable way and be detectable as artificially generated or manipulated, to the standard the article sets. That means the marking method must have been chosen, tested against foreseeable handling, implemented for each output type and delivery path, and documented with its limits. The provider should have completed the output inventory, the method testing, and the implementation before the deadline, with a buffer for remediation if testing finds a gap. A plan that schedules testing for the week before the deadline, with no time to fix a failure, is not a credible plan. The provider should record the implementation date, the version to which it applies, and the test results, so that compliance on the date can be shown from the file.

What else applies on 2 December 2026?

The same date, 2 December 2026, is the application date for two new prohibitions added by Regulation (EU) 2026/1744: Article 5(1)(ba), concerning AI systems used to generate or manipulate non-consensual intimate material, and Article 5(1)(bb), concerning material within Directive 2011/93/EU on combating sexual abuse and sexual exploitation of children. Those prohibitions are described in this guide only in general terms, by subject matter, without graphic detail or examples involving identifiable people. Teams should check the consolidated Article 5 wording, including the consent and defence conditions in outline, and should ensure that no system or use within the prohibition is relied on after that date. The coincidence of dates does not merge the two matters: the marking transition and the new prohibitions are distinct legal events that happen to share a date, and each should be planned separately.

How should a provider plan the final weeks before the date?

Planning for the final period should include a freeze on changes that could affect marking, a final test run on the production configuration, a check that the marking survives the live delivery paths, and a communication to deployers and downstream users about what has changed. The provider should also check that the disclosure and marking records are complete, that the version history is current, and that the team that will handle questions after the date is briefed. A provider that discovers a gap in the final weeks should record the gap, assess whether it affects the marking duty, and take corrective action before the deadline, rather than launching with a known gap and a plan to fix it later. The final check should be signed off by a named owner, with the date and the evidence recorded, so that the position on 2 December 2026 is clear from the file.

Continue with Article 50 marking and detection for synthetic content.

What does this guide not decide?

This guide does not decide whether a particular system qualifies for the transition, whether its marking meets Article 50(2), or whether a use falls within the new Article 5 prohibitions. It does not assess a placing date, a marking method, or a prohibition analysis. Those questions depend on placing evidence, version history, marking testing, the system's function and use, and the consolidated legal text, including Regulation (EU) 2026/1744. A provider should confirm its transition and prohibition position with a competent authority or legal adviser, keep a dated plan to 2 December 2026, and treat the date as a hard deadline for the marking work it covers.

Frequently asked questions

Continue through this AI Act cluster

• Transparency duties for AI
• Article 50 marking and detection for synthetic content
• Prohibited practices under Article 5
• What changes on 2 December 2026
• EU AI Act timeline
• When must people be told they are interacting with AI?