A date is circulating in commentary on the EU AI Act: 2 December 2026, described as the day a labelling or marking obligation for AI-generated content takes effect. That description is wrong. No transparency obligation enters into application on 2 December 2026. The obligations in Article 50, including the Article 50(2) duty to mark synthetic content in a machine-readable way, have applied since 2 August 2026, and the Digital Omnibus on AI (Regulation (EU) 2026/1744, in force 27 July 2026) left that date untouched.
For the transparency obligations and the prohibited practices, two things do happen on 2 December 2026, and both are worth getting precisely right. A transitional concession for older systems runs out, and a set of new prohibitions enters into application. Neither is a labelling obligation beginning.
The obligation did not move. The grace period for older systems ends.
Article 50(2) requires providers of AI systems that generate synthetic audio, image, video or text content to ensure the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. That duty has bound providers since 2 August 2026.
The Digital Omnibus added a transitional rule for systems that were already on the market before that duty arrived. Article 111(4), inserted by Regulation (EU) 2026/1744, reads:
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.
Recital 38 of the amending regulation describes this as a four-month transitional. Read it narrowly, because it is narrow in three ways.
First, it covers Article 50(2) only. The other Article 50 duties, including the Article 50(1) duty to disclose that a person is interacting with an AI system and the Article 50(4) deployer duties for deep fakes and published text, received no transitional. They have applied to everyone in scope since 2 August 2026.
Second, it covers systems placed on the market before 2 August 2026. A system placed on the market on or after that date never had the benefit of the transitional and owed Article 50(2) compliance from day one.
Third, it is a duty to come into compliance, not a suspension of the obligation. The provision's own words are "shall take the necessary steps in order to comply." A provider of a pre-August system that has done nothing is not exercising a grace period. It is running out one.
So the accurate statement of 2 December 2026 for transparency purposes is this: it is the day the last lawful reason for unmarked synthetic content from an in-scope provider expires. From 3 December 2026 there is no cohort of AI systems generating synthetic content on the EU market for which Article 50(2) does not bind.
A scope discrepancy worth knowing about
The Commission's Article 50 Guidelines (C(2026) 5054 final of 20 July 2026, content approved, formal adoption pending, and non-binding by their own paragraph 5) describe this transitional at paragraph 153 as covering systems "placed on the market or put into service" before 2 August 2026. The inserted Article 111(4) says "placed on the market" only. Those are distinct legal concepts under the Act: placing on the market is the first making available on the Union market, putting into service is first use for intended purpose.
For most providers the difference will not matter. For a provider whose system was put into service before 2 August 2026 without a placing on the market, it might. The binding text is Article 111(4), and this analysis cites it. Where the distinction is live for your system, that is a question for counsel, not for a guidance paragraph that is broader than the provision it describes.
What actually begins on 2 December 2026: new prohibitions
The Digital Omnibus inserted new prohibited practices into Article 5(1), and under Article 113(3)(a) as amended these enter into application on 2 December 2026.
Article 5(1)(ba) prohibits the placing on the market, putting into service or use of an AI system that generates or manipulates realistic images, videos, audio or similar material of an identifiable person's intimate parts, or of an identifiable person engaged in sexually explicit activities, without that person's freely given, specific, informed, unambiguous and explicit consent.
Article 5(1)(bb) prohibits the same acts for AI systems generating or manipulating material or performance within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU, the directive on combating child sexual abuse, subject to a narrow defence under national law.
Two accompanying paragraphs set the scope. Article 5(1a) limits the prohibition on placing on the market and putting into service to systems whose intended purpose is such generation, or whose design, training, architecture, capabilities or user-facing functionality make it a reasonably foreseeable and reproducible outcome without significant technical modification, where the system lacks reasonable and adequate safeguards to reliably prevent it and to correct observed or reported misuse. For deployers, the use prohibition bites where the system is used for the purpose of generating such material. Article 5(1b) carves out manipulation that does not increase the exposure of depicted intimate parts or alter the nature of depicted activities.
Note the structure of Article 5(1a)(a)(ii) for providers of general-purpose generation systems. The prohibition does not require intent. A system whose capabilities make the prohibited output reasonably foreseeable and reproducible, and which lacks adequate safeguards against it, is within the prohibition. The adequacy of safeguards, and the correction of reported misuse, are the operative questions, and both are questions a provider will be asked to evidence.
Article 5 prohibitions sit outside the risk-tier architecture. The Digital Omnibus deferred the high-risk regime (to 2 December 2027 for Annex III, 2 August 2028 for Annex I products), and deferred nothing here. A prohibited practice is unlawful regardless of risk classification, documentation or oversight. No record of careful conduct makes a prohibited system lawful.
What this asks of teams shipping agents
For a team deploying AI agents that produce synthetic content, three questions fall out of the above, in order.
Whether your provider marked the output. The Article 50(2) duty is the provider's, and after 2 December 2026 no provider on the EU market has a transitional to point to. A deployer building on a model or system that does not mark its output is building on a provider's open obligation.
Whether your own record shows what was generated. When the question arrives, from a counterparty, an auditor or a market surveillance authority, "the provider handles marking" is an answer about someone else's duty. What was generated in your runs, and whether it carried the required marking, is a fact about your deployment, and it is only answerable from a record made at the time.
Whether anything your agent generates could fall within the new prohibitions. For most enterprise deployments the honest answer is no by design. The provision rewards being able to show the design: the safeguards, and the handling of any reported misuse, are what Article 5(1a) turns on.
The dates in this piece, stated once
- 27 July 2026: Regulation (EU) 2026/1744 (Digital Omnibus on AI) enters into force.
- 2 August 2026: Article 50 transparency obligations apply, all paragraphs, all in-scope actors.
- 2 December 2026: Article 111(4) transitional for pre-2 August 2026 systems ends for Article 50(2); Article 5(1)(ba) and (bb) prohibitions, with Articles 5(1a) and 5(1b), enter into application.
- 2 December 2027: Annex III high-risk obligations apply.
- 2 August 2028: Annex I regulated-product high-risk obligations apply.
Sources: consolidated EU AI Act, CELEX 02024R1689-20260727; Regulation (EU) 2026/1744, ELI eur-lex.europa.eu/eli/reg/2026/1744/oj/eng; Commission Guidelines on Article 50, C(2026) 5054 final of 20 July 2026, non-binding, formal adoption pending.
Truveil records each run's events as they happen, and the engine version that scored them.
Frequently asked questions
Does an AI labelling obligation take effect on 2 December 2026?
No. The Article 50 transparency obligations, including machine-readable marking of synthetic content under Article 50(2), have applied since 2 August 2026. 2 December 2026 is the end of a transitional under Article 111(4) for systems placed on the market before 2 August 2026, not the start of a new obligation.
What is Article 111(4) of the EU AI Act?
A transitional inserted by the Digital Omnibus (Regulation (EU) 2026/1744). Providers of AI systems generating synthetic audio, image, video or text content that were placed on the market before 2 August 2026 must take the necessary steps to comply with Article 50(2) by 2 December 2026. It covers Article 50(2) only and systems placed on the market only.
What new prohibitions apply from 2 December 2026?
Article 5(1)(ba) prohibits AI systems that generate or manipulate realistic intimate or sexually explicit material of an identifiable person without that person's explicit consent. Article 5(1)(bb) prohibits generation or manipulation of material within the meaning of Article 2(c) and (e) of Directive 2011/93/EU. Articles 5(1a) and 5(1b) set the scope, including a safeguards test for providers of general-purpose generation systems.
Did the Digital Omnibus delay the EU AI Act's transparency rules?
No. The Omnibus deferred the high-risk regime, to 2 December 2027 for Annex III systems and 2 August 2028 for Annex I products. Article 50 transparency was not deferred and has applied since 2 August 2026.