EU AI Act transparency obligations (Article 50)
Article 50 applies to AI systems whatever their risk category: people must know when they talk to an AI, generated content must be detectable, and deepfakes must be disclosed.
Not legal advice. Sources: Regulation (EU) 2024/1689 (the AI Act).
The four duties
Article 50 sets four duties:
- AI that interacts with people (50(1), provider): designed so that people are informed they are interacting with an AI system, unless that is obvious.
- Synthetic content (50(2), provider): audio, image, video or text output is marked in a machine-readable format and detectable as artificially generated or manipulated, as far as technically feasible. Assistive editing that does not substantially change the input is excluded.
- Emotion recognition and biometric categorisation (50(3), deployer): the people exposed are informed.
- Deepfakes and public-interest text (50(4), deployer): disclose that image, audio or video content resembling real people, places or events was generated or manipulated; and disclose AI-generated text published to inform the public on matters of public interest, unless it was reviewed by a person who holds editorial responsibility.
How and when to inform
The information is given clearly, at the latest at the first interaction or exposure, and must meet accessibility requirements (Article 50(5)). For evidently artistic, creative, satirical or fictional work, the deepfake disclosure is limited so as not to hamper the work.
Codes of practice
The AI Office encourages codes of practice on detecting and labelling AI-generated content (Article 50(7)). Watermarks, metadata and content credentials are the usual technical means; check the current code and guidelines for what is expected.
Key dates
| Date | What applies |
|---|---|
| 1 August 2024 | The AI Act enters into force. |
| 2 February 2025 | Prohibited practices are banned; AI literacy duties start (Articles 4 and 5). |
| 2 August 2025 | Obligations for general-purpose AI models apply to new models (Articles 53 to 55); penalties apply. |
| 2 August 2026 | Most of the Act applies: high-risk systems in Annex III areas, transparency duties (Article 50), deployer obligations. |
| 2 August 2027 | High-risk rules apply to AI in products covered by Annex I law (Article 6(1)); general-purpose models placed on the market before 2 August 2025 must comply. |
Dates checked against Regulation (EU) 2024/1689 (the AI Act) on 17 November 2026. In November 2025 the European Commission proposed a "Digital Omnibus" that would postpone the high-risk obligations, linking them to the availability of standards (at the latest December 2027 for Annex III areas and August 2028 for Annex I products). Check whether it has been adopted, and its final dates, before relying on the high-risk dates shown here.
Questions
Does the transparency duty apply to minimal-risk systems?
Yes. Article 50 depends on what the system does (interacting with people, generating content), not on its risk category. A high-risk system can have transparency duties too.
Do I have to watermark AI-generated text?
Providers of systems that generate text must mark the output in a machine-readable way, as far as technically feasible (Article 50(2)). Deployers publishing AI text on matters of public interest must disclose it unless a person reviewed it and is editorially responsible.
More guides
Related free tools
Published ; awaiting review.