The 2 December 2026 marking deadline.

Reviewed

Short answer

Providers of generative AI systems placed on the EU market before 2 August 2026 must comply with the marking duty in Article 50(2) of the AI Act by 2 December 2026 (Article 111(4), added by Regulation (EU) 2026/1744). From then on their audio, image, video and text outputs must be marked in a machine-readable format and be detectable as AI-generated, as has applied since 2 August 2026 to systems placed on the market from that day.

Official sources: AI Act, Article 111(4) · Regulation (EU) 2026/1744

to the end of 2 December 2026, Central European Time (our reading of “by 2 December 2026”).

Which duty applies from when

Article 50 duties and the dates they apply from, with the 2 December 2026 transition
SituationDutyApplies from
Generative AI system, including a general-purpose AI system, placed on the EU market before 2 August 2026Mark outputs in a machine-readable format and make them detectable as AI-generated (Article 50(2))2 December 2026 (Article 111(4))
Generative AI system placed on the market on or after 2 August 2026The same duty2 August 2026 (Article 113)
AI system that talks to people directly, such as a chatbot, even one that also generates contentTell people they are interacting with an AI system (Article 50(1))2 August 2026; the transition covers only the marking duty (Commission guidelines, point 153)
Deployer publishing a deepfake, or AI-generated text on matters of public interestDisclose it (Article 50(4))2 August 2026, with no transition
Content generated before 2 August 2026No marking or labelling after the fact; text generated earlier but published from 2 August 2026 must be labelled where Article 50(4) appliesCommission guidelines, point 154

What “placed on the market” means

Placing on the market is “the first making available of an AI system or a general-purpose AI model on the Union market” (Article 3(9)), and making available means supplying it for distribution or use in the EU in the course of a commercial activity, “whether in return for payment or free of charge” (Article 3(10)). The date that counts is therefore when the system was first offered in the EU, paid or free. The Commission’s guidelines describe the rule as covering generative systems “placed on the market or put into service before 2 August 2026”.

  • Being outside the EU does not change this: the AI Act applies to providers placing AI systems on the EU market whether they are established in the EU or in a third country (Article 2(1)(a)).
  • The duty sits with the provider, the one who develops the system or has it developed and offers it under its own name or trademark (Article 3(3)); a business that only uses the tool is a deployer.
  • Article 50(2) does not apply where a system performs an assistive function for standard editing, does not substantially alter the input data or its semantics, or is authorised by law to detect, prevent, investigate or prosecute criminal offences.

Before 2 December 2026: a checklist for providers

  • List each generative feature you offer in the EU (image, video, audio, text) and the date it was first made available there.
  • Mark every output in a machine-readable format, with one technique or a combination, as long as the whole solution is effective, interoperable, robust and reliable as far as technically feasible (Article 50(2); guidelines, point 72).
  • Make the mark detectable: marking without available means of detection does not meet Article 50(2) (guidelines, point 70), and the result should be readable by people (point 75).
  • If your model provider or a third party marks the outputs for you, check that it covers your outputs; showing compliance stays your responsibility (guidelines, point 74).
  • Decide whether to sign the Code of Practice: its Section 1 asks signatories to mark images, audio and video with at least two layers, signed metadata and an imperceptible watermark, and to offer detection free of charge.
  • Keep marks intact in your own pipeline: the code asks signatories not to remove existing marks.
  • Write down which exception you rely on, if any, and why.
  • Keep the chatbot disclosure separate: Article 50(1) has applied since 2 August 2026.

What the Commission’s guidelines expect of a technical solution

The marking and detection solution as described in the Commission’s Article 50 guidelines
ElementWhat the guidelines say
MarkingMarks software can identify and extract without human intervention; visible labels may be added as a complement (point 71)
DetectionMeans of detection available to the people exposed to the content, with a human-readable result (points 75 and 77)
TechniquesWatermarks, metadata, cryptographic proof of provenance, logging, fingerprints, or a combination (recital 133; point 73)
Where in the chainAfter generation, in the model or during inference, or through an upstream or third-party solution (point 74)
QualityEffective, interoperable, robust and reliable as far as technically feasible, in line with the state of the art (points 80 to 83)
Code of PracticeAdherence to a code assessed as adequate can be used to show compliance with Article 50(2) and (5) (point 84)

After 2 December 2026

Once the date has passed, an existing generative system no longer has the benefit of Article 111(4). A breach of the Article 50 duties can be fined up to EUR 15 million or 3 % of total worldwide annual turnover for the preceding financial year, whichever is higher (Article 99(4)(g)); for SMEs, including start-ups, the lower of the two amounts applies (Article 99(6)), and since 2026 also for small mid-cap enterprises (Article 99(6a)). National market surveillance authorities enforce Article 50, and the AI Office does so for AI systems built on a general-purpose AI model by the same provider (guidelines, note 47).

Questions

Does the 2 December 2026 deadline apply to deepfake labels?
No. Article 111(4) covers only the provider’s marking duty in Article 50(2). The deployer’s duty to disclose deepfakes and AI-generated text on matters of public interest (Article 50(4)) has applied since 2 August 2026.
Do images generated before 2 August 2026 have to be marked now?
No, according to the Commission’s guidelines: outputs generated before 2 August 2026 do not need to be marked or labelled retroactively. Text generated earlier but published on or after that date must be labelled where Article 50(4) applies.
Our chatbot also creates images. Which date applies?
Both, for different duties. The images must be marked by 2 December 2026 if the system was on the market before 2 August 2026, but telling users they are talking to an AI has been required since 2 August 2026 (guidelines, point 153).

Sources

Last reviewed by Volkan Çatak, Founder, Proofcan.

Not legal advice. This page explains official texts in plain words and links each one; check the sources, or a lawyer, before relying on it for a decision.