The 2 December 2026 deadline for marking AI-generated content

Published 12 October 2026

Article 50 of the EU AI Act has applied since 2 August 2026. One part of it has a later date. Providers of generative AI systems that were already on the market before 2 August 2026 have until 2 December 2026 to mark what those systems generate. This page sets out what that date covers, from the text of the Act and published summaries of it, and the technical work it implies.

What has to be in place by 2 December

Article 50(2) says that providers of AI systems that generate synthetic audio, image, video or text must ensure the outputs are marked in a machine-readable format and are detectable as artificially generated or manipulated (Article 50 of the AI Act (artificialintelligenceact.eu)). General-purpose AI systems are included.

The same paragraph says the technical solutions should be effective, interoperable, robust and reliable, as far as that is technically feasible, taking into account the type of content, the cost and the generally acknowledged state of the art.

There is an exception: the duty does not apply where a system performs an assistive function for standard editing, or does not substantially alter the input data or its meaning.

Which systems the date covers

Cooley's summary of the rules describes the transitional relief as narrow (Cooley, "EU AI Act: Transparency Obligations Take Effect 2 August 2026" (3 August 2026)):

It applies only to the marking and detection duty in Article 50(2).
It applies only to generative AI systems already on the market before 2 August 2026. Their providers have until 2 December 2026.
Every other in-scope obligation applied from 2 August 2026, whenever the system was placed on the market.
Content generated and published before 2 August 2026 does not need to be labelled retroactively.

So a generative system launched after 2 August 2026 does not get the extra four months. The European Commission adopted guidelines on Article 50 on 20 July 2026, and Cooley reports fines of up to 15 million euros or 3% of worldwide annual turnover, whichever is higher.

What already applied on 2 August

Article 50(1): AI systems that interact directly with people, such as chatbots and voice assistants, must tell people they are dealing with an AI system, unless that is obvious from the context.
Article 50(4): deployers of systems that produce deep fakes must disclose that the content is artificially generated or manipulated, with lighter rules for evidently artistic, satirical or fictional work.
Article 50(4): deployers who publish AI-generated text to inform the public on matters of public interest must disclose it, unless the text has had human review and someone holds editorial responsibility for it.

A technical checklist for the marking duty

None of this decides whether the duty applies to you, or whether a given measure meets it. That is a legal question for you and your advisers. These are the engineering steps the duty points to.

List every output type. Write down whether your system produces images, audio, video, text, or several of these. Each needs its own marking method.
Pick a machine-readable mark for each. For image, audio and video files, the open standard is a C2PA manifest, also called Content Credentials. It is signed metadata that states how the file was made. For images, add an invisible watermark as well, because metadata is easy to strip. Text is harder to mark reliably, and the approaches differ by provider.
Mark at the point of generation. Add the marks before the output leaves your system, so every copy carries them.
Test that the marks survive your own pipeline. Content delivery networks, image resizers and content management systems often rewrite files and drop metadata. Check a file as your users actually receive it, not only the file your model wrote.
Make the marks detectable. The Act asks for outputs to be detectable, not only marked. Point people to a tool that reads your marks, such as a C2PA reader or a verify page.
Keep a record. Keep a log of what was marked and when, so you can show your marking was running.
Read the Code of Practice. The AI Office's voluntary Code of Practice on Transparency of AI-Generated Content sets out icons for labelling AI-generated content (European Commission, EU icons for labelling AI-generated content). Cooley notes that signatories are treated more favourably in enforcement.

Where Tellmark fits

Tellmark covers steps 2, 3, 5 and 6 for images. It adds a signed C2PA manifest and an invisible TrustMark watermark to each AI-generated image, logs AI-generated text with a visible label, keeps a marking log, and offers a free checker and a per-customer verify page. You remain the provider and remain responsible for your obligations. Tellmark supplies technical measures, not a legal outcome, and what marking cannot do sets out its limits.

Sources

Related guides

Describes technical measures. Not a legal assessment.