EN
Log in Get started
Guides

The EU AI Act and your AI presenter: what a business has to label (2026)

What Article 50 asks of a business that posts videos with an AI-generated person: who has to label them, what the label has to look like, the dates, and a checklist.

Last checked 15 October 2026. General information, not legal advice: check with whoever signs off your compliance.

The short version

Since 2 August 2026, Article 50 of the EU AI Act requires whoever publishes a realistic video of a person who isn't real to say that it's AI-generated. A presenter you made up for your business counts, even though nobody is meant to be fooled. The label has to be visible from the start: a hidden watermark isn't enough.

When it applies

Article 50 applies from 2 August 2026. The Digital Omnibus on AI (Regulation (EU) 2026/1744) postponed the rules for high-risk AI systems, but not Article 50.

There is one grace period, and it's for the tools rather than for you: tools that make AI video and were on the market before 2 August 2026 have until 2 December 2026 to mark what they make in a machine-readable way. Videos made before 2 August 2026 don't need labels added after the fact.

Who has to label

The Act splits the duty in two. The provider of the tool that makes the video has to mark it in a way software can read. The deployer, whoever uses the tool to publish the video, has to tell people it's AI.

Law firms reading the Commission's guidelines (final, 20 July 2026) say the deployer is whoever decides how the AI is used. When an agency runs the presenter for a client, that is usually the agency, not the client who commissioned it.

Invented people count

The guidelines read “existing persons” broadly: a photorealistic portrait of someone who doesn't exist still counts as a deep fake, and no intent to deceive is needed.

Marketing gets no lighter treatment. The softer labelling allowed for artistic or satirical work doesn't cover content that is purely informative or commercial. And an AI image that makes a product or a property look better than it is counts as well.

What the label has to look like

Clear and noticeable at the latest when someone first sees the video: on screen as it starts, not only at the end or in the description. Machine-readable marking alone isn't enough, and you can't rely on the tool's invisible watermark.

In practice: an “AI-generated” label on screen from the first second, in the video's own language, plus the platform's AI setting when it's posted and a line in the caption.

Comments and messages

If your presenter ever answers comments or messages, people have to be told they're talking to an AI (Article 50(1)).

Fines

Breaking Article 50 can cost up to 15 million euros or 3% of worldwide yearly turnover. For small and medium-sized businesses and start-ups, the lower of the two applies.

A checklist

☐ An on-screen label that says AI, from the first second to the last, in the video's language.

☐ The platform's own AI setting switched on for every post: TikTok's AI-generated content flag, YouTube's altered or synthetic content, Instagram's AI info.

☐ On Instagram, the account's AI-generated profile label switched on.

☐ “AI-generated” in the caption too, where it can't be cropped out.

☐ Machine-readable marking inside the file (IPTC metadata, and signed Content Credentials where your tool has them).

☐ A record of each video: what it said, where that came from, who approved it and when.

☐ If you work through an agency: agree in writing who labels what.

Sources

Have these made and posted for you, every day

ViralBrand gives your business its own AI influencer. It learns what you do from your website and your team's notes, makes a short video every day with its own face and voice, and posts it to your TikTok, Instagram, YouTube and Facebook once you tap approve. Every video is labelled as AI.

Get started
From €49 a month for 3 videos a week, or €149 for one a day. No contract.

More free help