EU AI Act Art. 50 — Transparency obligations for providers and deployers of certain AI systems
Article 50 is the rulebook for lower-risk AI that still interacts with people: chatbots must say they are AI, and synthetic content (deepfakes, AI-generated audio) must be labelled. It is the Act's most visible obligation for everyday products.
At a glance
What this article requires
- Providers must design chatbots and other interactive AI so users know they are talking to AI, unless it is obvious.
- AI-generated audio, video, text, or images that look real must be disclosed as artificially generated.
- Deepfake content must be labelled as such, with limited exceptions for law enforcement, art, and satire.
- Emotion-recognition and biometric-categorisation systems must tell affected people they are being processed.
- All Article 50 transparency obligations — chatbot disclosure, synthetic-content labelling, and emotion/biometric notification — apply from 2 August 2026.
Scope
Who this applies to
Providers (who must design disclosure in) and deployers (who must display it) of chatbots, synthetic content generators, deepfakes, and emotion or biometric systems — across all risk tiers below high-risk.
Obligations
What you must actually do
Label the conversation
If your chatbot or voice assistant could be mistaken for a human, disclose the AI nature at the start of interaction — and keep the disclosure when the assistant seems capable of emotional manipulation.
Mark synthetic content
AI-generated or manipulated media that could pass for real must be marked in a machine-readable way and disclosed to viewers — covering images, video, audio, and text.
Notify for emotion and biometrics
People subject to emotion recognition or biometric categorisation must be informed that they are being processed. This is separate from (and in addition to) high-risk Annex III duties.
Action plan
Practical first steps
- 1
Audit every customer-facing AI for disclosure gaps: chatbots, voice agents, avatar videos, generated marketing images.
- 2
Add machine-readable marking to synthetic media in your pipeline (watermarks, metadata).
- 3
Keep an exemption log for artistic, satirical, or fiction uses where you rely on Art. 50 exceptions.
- 4
Ship disclosure in the UX before you ship the feature — it is a design requirement, not a footer footnote.
Penalty exposure
Transparency violations are generally in the middle tier: up to €15 million or 3% of global annual turnover (the €7.5M/1% tier applies for supplying incorrect, incomplete, or misleading information to authorities).
FAQ
Questions about Art. 50
Does my customer-service chatbot really need to say it's a bot?
Yes, unless it is obvious from the context. The EU AI Act requires disclosure when a user interacts with AI and could reasonably believe it is human — this is one of the most commonly enforced transparency rules in practice.
Are AI-generated marketing images deepfakes that need labels?
If an AI-generated image or video could reasonably be mistaken for real footage of real people or events, it must be disclosed as synthetic. Clearly stylised or obviously artificial content may not need it — judge by deception risk.
Sources
Citations & further reading
Related
More article explainers
Wondering which articles apply to your AI?
Describe your system in the free Risk Scanner and get a preliminary risk read with the obligations that likely apply — in seconds.
Check my use casePreliminary EU AI Act clarity summary. Not legal advice.