EU AI Act glossary

Transparency obligations (Article 50)

Article 50 of the EU AI Act requires people to be told when they are dealing with AI, and AI-generated content and deepfakes to be marked or disclosed, from 02/08/2026.

In plain English

Article 50 is the part of the AI Act that most small builds actually hit, because it applies to ordinary chatbots and content generators, not just high-risk systems. It applies from 02/08/2026.

Providers must design AI systems that interact directly with people so those people are told they are dealing with AI, unless that is obvious to a reasonably well-informed person. Providers of systems that generate images, audio, video or text (including a chatbot's replies) must make sure the output is marked in a machine-readable way and can be detected as AI-generated. Relying on the AI model's own marking is fine only if it is good enough, because the responsibility stays with the provider. Standard editing help, such as fixing grammar, and changes that don't substantially alter what the user supplied don't need marking. For systems already on the market before 02/08/2026, the Digital Omnibus gives until 02/12/2026 for that marking. (The law says "placed on the market"; Commission guidance reads it as also covering systems already in use.)

Deployers must tell people when they use emotion recognition or biometric categorisation on them, although some of these uses are banned outright, such as inferring emotions at work or in education (Article 5). They must disclose deepfakes: AI-generated or altered images, audio or video that resemble real people, objects, places or events and would falsely appear genuine (obviously artistic or satirical work only needs a lighter label). And they must disclose AI-generated text published to inform the public on matters of public interest, unless a person has genuinely reviewed it and someone takes editorial responsibility. Commission guidance says that review must at least include fact-checking; a spell-check isn't enough, and if AI changes the text after the review, it needs the label again.

The information has to be clear and distinguishable, accessible, and given at the latest at the first interaction or exposure. Breaches can be fined up to €15 million or 3% of worldwide turnover, whichever is higher. For SMEs and small mid-cap companies, it is whichever is lower.

What it means for your build

  • Chatbots, voice agents and AI agents that email or message people: put an AI disclosure at the start of the conversation by default.
  • Anything that generates text, images, video or audio: check whether the model you use adds machine-readable marking (watermarks, or metadata such as C2PA content credentials) and make sure it is switched on. If it doesn't, or its marking isn't good enough, the system's provider must add marking.
  • Give the client ready-made disclosure wording for deepfakes and AI-written public content, because those duties fall on them as deployer.
  • Separately, from 02/12/2026 it's banned in the EU to supply an AI system built to create realistic sexual images, video or audio of a real, identifiable person without their explicit consent, or child sexual abuse material. It's also banned to supply one where that output is a foreseeable, repeatable result and it lacks adequate safeguards. If your build generates images, video or audio, put strong safeguards in place and test that they work.

Sources

Related terms

Which of these apply to your build?

Answer about a dozen plain-English questions and get a checklist for the website, AI agent or automation you are building.

Check your build

Guidance, not legal advice. Last checked 13/11/2026. · How we check