EU AI Act Article 50: transparency obligations
Applies from 2026-08-02 · Last reviewed 2026-09-10 · Current
Article 50 of the EU AI Act imposes transparency duties. Providers must mark synthetic image, audio and video outputs so they are machine-detectably artificial. Deployers must disclose deepfakes — realistic AI-generated or manipulated content resembling real persons, objects, places or events — and must disclose AI-generated text informing the public on matters of public interest unless a human takes editorial responsibility. These obligations apply from 2 August 2026.
Who is affected
- Marketing teams acting as deployers of generative AI systems in the EU.
- Agencies producing campaign assets with generative components.
- Providers of generative tools whose outputs must be machine-marked.
What triggers the rule
- Generating or manipulating image, audio or video that resembles real persons, objects, places or events and would falsely appear authentic.
- Publishing AI-generated text to inform the public on matters of public interest.
What does not trigger it / exceptions
- AI-generated text that went through human editorial review where a natural person takes editorial responsibility.
- Artistic, satirical or fictional deepfake work — disclosure still applies, but in an appropriate manner that does not hamper the work.
Scenario table
| Scenario | Outcome | Why |
|---|---|---|
| Synthetic voice-over on real product footage | Disclosure recommended | Not automatically a deepfake unless it resembles a real person’s voice or falsely appears authentic; platform policies and client standards usually recommend disclosure. |
| AI video of a real-looking person endorsing a product | Disclosure required | Article 50(4) deepfake disclosure; also a consent and rights review. |
| AI-generated news-style article, no human review | Disclosure required | Article 50(5) public-interest text disclosure applies. |
Required action
- Identify assets containing realistic synthetic image, audio or video.
- Disclose the artificial generation or manipulation clearly and distinguishably.
- Do not rely on tool-side metadata alone — record the decision yourself.
- Route identifiable-person and consent questions to human review.
Practical marketing examples
- A campaign video with an AI-generated crowd scene that looks filmed on location — disclose the synthetic scene.
- A cloned voice of a celebrity — do not publish; human review and rights clearance required.
Relation to platform policy
EU AI Act Article 50(4) requires deployers to disclose realistic AI-generated or manipulated image, audio and video resembling real persons, objects, places or events. Use the platform’s native label and keep your own evidence record.
Run the free check — result, wording and sources in under a minute.
Primary sources
Frequently asked questions
Does Article 50 apply to advertising?
Yes where its triggers are met — realistic synthetic image, audio or video resembling real persons, objects, places or events must be disclosed regardless of whether the use is commercial.
Is a caption enough?
The law requires disclosure that is clear and distinguishable. Platform-native labels are the most reliable mechanism on social platforms; captions and on-asset wording can complement them.
Related pages
Methodology: Disclara answers are generated from a versioned rule set citing primary sources, reviewed by humans. Information on this page is not legal advice. See sources & methodology and rule changelog.