VITNADoes the December deadline
apply to you?
If your generative AI product was already on the Union market before 2 August 2026, you have until 2 December 2026 to mark its output in a machine-readable format. Systems that launched on or after 2 August 2026 are already subject. A few questions tell you which side of that line you are on. No account, no email, and the result is a shareable link.
This tool determines what applies. It never claims to determine whether you have met it, and it does not score you.
Do people in the EU use your product, or does its output reach them?
The Act follows the output, not your address. If people in the EU see, receive or rely on what your system produces, it reaches you.
Machine-readable marking, for systems already on the market
Providers of systems generating synthetic audio, image, video or text that were placed on the Union market before 2 August 2026 have until 2 December 2026 to comply with the marking requirement. Systems placed on the market on or after 2 August 2026 have been subject since that date, with no transitional period.
Up to EUR 15,000,000 or 3% of total worldwide annual turnover, whichever is higher (Article 99(4)).
Two new prohibited practices
The Digital Omnibus added prohibitions on AI systems that generate or manipulate non-consensual intimate imagery of identifiable people, and on AI generation or manipulation of child sexual abuse material. Both apply from 2 December 2026 and reach providers and deployers alike.
Up to EUR 35,000,000 or 7% of total worldwide annual turnover, whichever is higher (Article 99(3)).
Who the marking requirement reaches
Article 50(2) falls on providers — whoever places a system on the market or puts it into service under their own name or trademark. That includes companies building on another firm's model: if the product carries your brand, you are its provider, and the fact that the model underneath is someone else's does not move the obligation to them. It applies whether the product is paid or free, and it reaches providers established outside the Union whenever the output is used inside it.
What machine-readable marking means
The output itself has to carry something a machine can detect — a watermark, embedded metadata, or content credentials — so that the content can be identified as artificially generated or manipulated without a human judging it by eye. A visible “made with AI” label is not the same thing: the requirement is machine detectability. Article 50(2) also sets a quality bar. The technical solution has to be effective, interoperable, robust and reliable as far as is technically feasible, taking into account the specificities of the content type, the cost of implementation, and the generally acknowledged state of the art. In practice, marking that does not survive ordinary handling of a file is unlikely to meet that bar.
The two dates, and why they differ
Article 50 has applied since 2 August 2026. The Digital Omnibus did not change that. What it added was a transitional provision — a new Article 111(4) — for providers whose generative systems were already on the Union market before that date: they have until 2 December 2026 to bring the marking into line. A system that first reached the market on or after 2 August 2026 got no such period and has been subject from that day. The trigger is market placement — the first time the system was made available in the Union — not the date you built it, and not each subsequent release.
The prohibitions arriving on the same day
The same date brings two additions to Article 5, the Act's list of prohibited practices: AI systems that generate or manipulate realistic intimate imagery of an identifiable person without their explicit consent, and AI generation or manipulation of child sexual abuse material within the meaning of Directive 2011/93/EU. These are prohibitions rather than transparency duties — no amount of marking or disclosure makes the practice permitted — and they carry the Act's highest penalty tier. They reach deployers as well as providers: using such a system for that purpose is caught alongside placing one on the market.
Sources: Regulation (EU) 2024/1689 (AI Act), OJ L, 2024/1689, 12.7.2024; Regulation (EU) 2026/1744 (Digital Omnibus on AI), OJ L, 2026/1744, 24.7.2026, in force 27 July 2026 — Article 1(7) (new Article 5 points), Article 1(39)(b) (new Article 111(4)), Article 1(40) (application dates).
What exactly happens on 2 December 2026?
Two things. Providers of generative AI systems that were already on the Union market before 2 August 2026 reach the end of a four-month transitional period for Article 50(2) machine-readable marking. And two new prohibited practices, added to Article 5 by the Digital Omnibus, become applicable. Those are the only obligations in the amended Act carrying that date.
Didn't the Digital Omnibus delay the transparency obligations?
No. Article 50 has applied since 2 August 2026 and that did not change. What the Omnibus added is a transitional provision — Article 111(4) — giving providers whose systems were already on the market a further four months for the Article 50(2) marking duty specifically. The rest of Article 50, including chatbot disclosure and deepfake labelling, had no transition at all.
My system launched in September 2026. Do I have until December?
No. The transitional period is only for systems placed on the Union market before 2 August 2026. Anything that reached the market on or after that date has been subject to Article 50(2) since 2 August 2026.
What does machine-readable marking actually mean?
The output has to carry a signal another system can read — watermarking, embedded metadata, or content credentials — so the content is detectable as artificially generated without a person judging it by eye. Article 50(2) requires the technique to be effective, interoperable, robust and reliable as far as is technically feasible, judged against the generally acknowledged state of the art. A visible label alone does not satisfy it.
We build on someone else's model. Are we the provider?
If you place the product on the market or put it into service under your own name or trademark, you are its provider for these purposes, whoever trained the model underneath. That is the definition in Article 3(3).
Is there an exemption for editing tools?
Article 50(2) does not apply to the extent that a system performs an assistive function for standard editing, or does not substantially alter the input data provided by the deployer or its semantics. It is measured against what the system does to the input, so it narrows the obligation rather than removing it from a product that also generates content.
We are not established in the EU. Does this reach us?
It can. The Act reaches providers outside the Union whenever the output of the system is used in the Union. Physical presence in the EU is not required.
What is the next deadline after this one?
Obligations for high-risk systems under Annex III apply from 2 December 2027, and for high-risk systems embedded in Annex I products from 2 August 2028. Both dates are unconditional in the adopted text.
This page covers the 2 December 2026 obligations only. For the Article 50 transparency obligations that have applied since 2 August 2026 — telling people they are talking to an AI, emotion-recognition notice, deepfake and AI-text disclosure — use the Article 50 self-check.