Telling people they are talking to an AI: chatbot disclosure under Article 50

Published 12 October 2026

If your product lets people talk to an AI system, through a chat window, a voice assistant or an agent that answers email, Article 50(1) of the EU AI Act is the part that applies to it. It has applied since 2 August 2026, with no transitional period (Cooley, "EU AI Act: Transparency Obligations Take Effect 2 August 2026" (3 August 2026)).

What the text asks

Providers must design AI systems that interact directly with people so that those people are informed they are interacting with an AI system (Article 50 of the AI Act (artificialintelligenceact.eu)). It is a duty on the provider, the business that builds the system and puts it on the market.

Article 50(5) sets how: the information must be clear and distinguishable, given at the latest at the time of the first interaction, and must meet the applicable accessibility requirements.

When it is "obvious"

The duty does not apply where it is obvious from the point of view of a reasonably well-informed, observant and circumspect person that they are dealing with AI, taking the circumstances and context into account. Relying on this exception is a judgement call. A clear disclosure costs one line, so most products simply show it. There is also an exception for systems authorised by law to detect or investigate crime.

Showing it well

These are design patterns, not requirements from the text. The text asks only that the information be clear, distinguishable, timely and accessible.

Say it before or with the first reply. Put a line such as "You're chatting with an AI assistant." in the chat window's header or at the top of the conversation, before the person has to ask.
Keep it visible. A line that scrolls away after the first message is easy to miss for someone who joins a long thread later. A persistent label in the chat header avoids that.
Make it distinguishable. Present the line as interface text, not as part of the assistant's own reply. Give it enough contrast to read.
Make it accessible. Use real text, not an image of text, so screen readers announce it. For voice, say it at the start of the call.
Cover every channel. If the same assistant answers by email, messaging or phone, each channel needs its own disclosure.

The AI Office's voluntary Code of Practice includes an icon for interactive AI systems, which can sit beside the text (European Commission, EU icons for labelling AI-generated content).

Generated content is a separate duty

If the assistant also generates images, audio or video, those outputs fall under Article 50(2) as well, which asks for machine-readable marking rather than a visible message. See Article 50: marking and labelling AI-generated content.

Where Tellmark fits

Tellmark's dashboard produces a one-line chatbot disclosure, with a visible text label and an optional icon, as a snippet to paste into your chat interface. It is a convenience for showing the line consistently. Whether your interface meets Article 50(1) is for you and your advisers to judge.

Sources

Related guides

Describes technical measures. Not a legal assessment.