Transparency duties: chatbots, generated content and deepfakes

Article 50 sets transparency duties that apply whatever the risk tier.

What providers must do

• AI that interacts with people must make it clear they're talking to AI, unless it's obvious. Has applied since 2 August 2026.
• AI that generates audio, images, video or text must mark the output in a machine-readable way so it can be detected as AI-generated. This has applied since 2 August 2026; systems already on the market before that date have until 2 December 2026.

What deployers must do (since 2 August 2026)

• If you use emotion recognition or biometric categorisation, tell the people exposed to it.
• If you publish a deepfake (realistic image, audio or video of real people, places or events), disclose that it's AI-generated. Clearly artistic, satirical or fictional work only needs a light-touch disclosure.
• If you publish AI-generated text on matters of public interest, disclose it, unless a person reviewed it and takes editorial responsibility.

How to do it well

1. 1
   Tell people at the start of the interaction, not hidden in terms and conditions.
2. 2
   Use plain words: "You're chatting with an AI assistant."
3. 3
   For marking, use an established standard such as C2PA content credentials or watermarking your provider supports.
4. 4
   Keep a record of how and where you disclose. That's your evidence.