tattletagArticle 50 is the part of the EU AI Act (Regulation 2024/1689) that deals with transparency: people should know when they talk to a machine or look at something a machine made. It applies from 2 August 2026.
The five paragraphs in one line each
| Paragraph | Who | What they must do |
|---|---|---|
| 50(1) | Providers of AI systems that interact with people (chatbots, voice agents) | Tell people they are dealing with AI, unless it is obvious |
| 50(2) | Providers of generative AI | Mark output in a machine-readable way so it can be detected as AI-made (watermarks, metadata) |
| 50(3) | Deployers of emotion recognition or biometric categorisation | Inform the people exposed to it |
| 50(4) | Deployers who publish deepfakes or AI-generated text on matters of public interest | Disclose that the content is AI-generated or manipulated |
| 50(5) | Everyone above | Make the information clear and distinguishable, at the latest at first exposure, and accessible |
A provider builds or sells the AI system (OpenAI, Google, Midjourney). A deployer uses it in their own work: a newsroom, a brand, an agency, a blogger with a professional site. Purely personal, non-professional use is outside the AI Act.
Deepfakes: always labelled
A deepfake, in the Act’s words, is AI-generated or manipulated image, audio or video that resembles real people, objects, places or events and would falsely appear authentic. If you publish one, you must disclose it. Human editing does not change that: a deepfake a person polished is still a deepfake.
There is a lighter rule for work that is evidently artistic, creative, satirical or fictional. There, the disclosure must not spoil the work; a note next to it or a small icon is enough.
AI text: only on matters of public interest
AI-generated text must be disclosed when it is published to inform the public on matters of public interest: news, politics, health, public safety. Two conditions together lift the duty:
- the text went through human review or editorial control, and
- a person or company holds editorial responsibility for it.
A newsroom whose editor checks every AI-drafted article meets both. A site that publishes unchecked AI news does not. Product descriptions, recipes or a personal travel blog are rarely “matters of public interest” in the Act’s sense, though platform rules and honesty may still call for a label.
On Tattletag, level 3 (AI-made, human-checked) describes text that went through review; level 4 (AI-generated) describes text that did not.
What the label must look like
The Act itself only says “clear and distinguishable” at first exposure. The details come from the Code of Practice on transparency of AI-generated content, published by the Commission on 10 June 2026 and assessed as adequate by the Commission and the AI Board. Its labelling section (for deployers) says, in short:
- use the EU icon or an equivalent label with “AI” as the main visual element;
- on images and video, place it on the content, visible at first sight, without covering it;
- in video, show it at the start, and where possible at intervals and after ad breaks;
- for audio, say it at the start;
- for text, put it above the text, near the headline, or in the colophon;
- make it accessible: alt text, contrast, screen readers.
The Commission’s guidelines on article 50, approved in July 2026, explain the terms. Neither the Code nor the guidelines are law; deployers who do not sign the Code must show they meet article 50 in another way.
The official EU icons
The AI Office published three icons: a plain “AI” icon, “AI + GENERATED” and “AI + MODIFIED”, each in black, white and two 50% transparent versions. They are on the Commission’s icon page, free for everyone to use without attribution. The Commission adds that using them does not establish legal compliance by itself.
A Tattletag badge can sit next to an EU icon: the icon for the legal label on the content, the badge for the reader who wants to know more, with a link that explains the level.
Fines
Breaches of article 50 can cost up to €15 million or 3% of worldwide annual turnover, whichever is higher; for SMEs and start-ups, whichever is lower (article 99).
What to do now
- List what you publish that AI made or changed: images, video, audio, text.
- Mark the deepfakes and put a visible label on them.
- Check whether AI-written public-interest text goes through human review and who holds editorial responsibility. Write that down.
- Add a short note on your site saying how you use AI and what your labels mean. The Code asks signatories to do this, and readers like it.
- Pick a badge for everything else, so readers do not have to guess.
