What is a violation of the AI disclosure obligation?

A violation of the disclosure obligation under Art. 50 EU AI Act exists in practice when an AI system – such as a chat or voice widget – interacts directly with a natural person without that person being informed in a clear, distinguishable and accessible way that they are communicating with an AI.

Typical practical examples: a chat widget labelled only with a first name such as "Anna", with no further indication of AI. Or an AI notice that appears only in the legal notice (imprint) but not at the chat widget itself – where the interaction actually takes place. A notice that flashes up briefly but disappears automatically after a few seconds also usually does not reliably meet the requirement in practice.

The common denominator of such cases: the notice either does not exist at all, or it is placed or designed in such a way that an average user does not notice it at the moment of interaction.

Violations of Art. 50 fall under the fine regime of Art. 99 AI Act. You will find details on the specific fine framework in the entry Which fines can be imposed for AI Act violations?

Important: whether a violation actually exists in an individual case and how it is to be assessed depends on the specific design of the widget concerned – a blanket assessment cannot responsibly be made here. If you want to check for yourself whether your own website might be affected, you will find a practical self-test as well as example wordings for correct disclosure in the full guide to the chatbot disclosure obligation.