From December 2, 2026, generative AI systems placed on the EU market before August 2 must mark their output in a machine-readable way. That is less than a month away, and it is easy to miss: many teams heard that the Digital Omnibus postponed the AI Act and assumed they had time. The delay mainly covers high-risk systems. It did not cancel the transparency rules.
What changed in July
Regulation (EU) 2026/1744 of July 8, 2026, the Digital Omnibus on AI, moved the high-risk deadlines: for Annex III systems, to December 2, 2027. Article 50 on transparency has applied since August 2, 2026. Only one point got extra time: new Article 111(4) gives providers of systems released before August 2 until December 2, 2026 to comply with Article 50(2). A system placed on the market on or after August 2 has no such grace period.
What Article 50(2) requires
Providers of systems that generate synthetic audio, images, video or text must ensure the output is marked in a machine-readable format and detectable as artificially generated or manipulated. The solution has to be effective, interoperable, robust and reliable as far as technically feasible. The exception covers assistive functions for standard editing and systems that do not substantially alter the input or its meaning.
What already applies
- People are told they are dealing with AI unless that is obvious from the context: chats, assistants, voice bots. This is Article 50(1).
- People exposed to emotion recognition or biometric categorisation are informed: Article 50(3).
- Deepfakes are disclosed. So is text on matters of public interest, unless a person reviewed it and someone holds editorial responsibility: Article 50(4).
- The information is given clearly, by the first interaction at the latest, and meets accessibility requirements: Article 50(5).
Does this reach companies outside the EU?
It can. The AI Act covers providers placing systems on the EU market wherever they are based, and cases where the system's output is used in the EU. Whether you count as a provider or a deployer, especially when your product is built on someone else's model, is for your lawyer to decide. Your role determines which paragraphs apply.
What to check before December 2
- List your AI features: what they generate, for whom, on which model, and when they launched.
- Agree your role for each feature with your lawyer.
- Ask your model vendor what marking it provides and in what form.
- Test whether the marking survives your pipeline. Compression, resizing, transcoding and CDNs can strip metadata. Follow a real file from the model to the user.
- Close the gaps. There is no single technology: the Code of Practice accepts that no one method meets every requirement yet. Methods are usually combined, for example C2PA metadata with watermarks.
- Check the AI notice in chats, assistants and voice flows: it is already mandatory.
- Add a marking check to CI so the next update does not break it.
The Code of Practice and fines
The final Code of Practice on marking and labelling AI content was published on June 10, 2026. It is voluntary and will likely serve as a reference point for showing compliance, but the Article 50 obligations themselves are binding. Article 99 sets fines of up to EUR 15 million or up to 3% of worldwide annual turnover, whichever is higher. For SMEs, the lower of the two applies.
This article explains the requirements and does not replace legal advice. We handle the engineering side: AI feature inventory, notices, content marking, logs and technical documentation.
Sources: Regulation (EU) 2026/1744, Article 50, Article 2, Article 99, European Commission on the Code of Practice, Jones Day, Paul Weiss, DLA Piper.