Skip to content

On 2 December the extension for labelling AI-generated content ends: what falls to you and what to your provider

On 2 December the extension for labelling AI-generated content ends: what falls to you and what to your provider

On 24 July 2026 Regulation (EU) 2026/1744 was published, the so-called digital Omnibus on AI, which amends the European Artificial Intelligence Regulation. It postponed quite a few things, mostly the high-risk ones. Article 50, on transparency, was not postponed: it applies from 2 August 2026. With a single exception, and that one has a date: 2 December 2026.

What expires on 2 December

Article 50(2) requires providers of AI systems that generate audio, image, video or text to make sure that what they produce is marked in a machine-readable way, so that it can be detected as generated or manipulated by an AI. In practice, things like watermarks or metadata that travel with the file.

The Omnibus gave a margin only to tools that were already on the market before 2 August. The official text says that those providers “shall take the necessary measures to comply with Article 50(2) by 2 December 2026 at the latest”. From that day on, the margin is over for everyone.

What isn't yours

If you use a tool to generate images, texts or videos, that technical marking is not yours: it belongs to whoever builds the tool. You don't have to install anything or add metadata by hand.

The exception is when you are the provider. If your company has developed a tool that generates content and offers it to others under its own name, this date is yours too.

What is yours, and since August

The rest of Article 50 already applies. Two examples that affect almost any company that publishes:

  • If you publish AI-generated images, audio or video that pass as real people, places or events, you have to say they are artificial.
  • If you publish AI-generated text to inform the public on matters of public interest, the same applies, unless a person has reviewed it and answers for it.

And if your website has an assistant that talks to your customers, whoever talks to it has to know they are talking to an AI.

We covered this in more detail on 2 August, in our piece on what Article 50 requires and who is liable for it.

Three things you can do before 2 December

  • Make the list. Note down which tools your company uses to generate content: images for social media, texts, videos, voices.
  • Ask each provider. Whether it marks what it generates and how. It is a short question, and the answer tells you a lot about how it takes compliance.
  • Review your notices. The ones that depend on you, the visible ones, have been required since August. If you don't have them yet, the December date isn't yours: yours has already passed.

What we'd rather not promise you

We don't know how it will be checked in practice that a tool marks what it generates, nor which technical solutions will end up being the usual ones. Nor do we tell you which tool to choose: it depends on what you do with it. What we do know is the date, and that it is written in the official text.

How we do it ourselves with our own content is in the transparency statement (in Spanish).

If you want to know what falls to you and what to your provider, the AI Act check tells you in nine questions, free and with no sign-up.