What the article actually says
Article 50(1): providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system. The test is what a person who is reasonably well-informed, observant and circumspect would understand, taking the circumstances and context of use into account. It is not a subjective test about your particular users.
Who carries it: the provider, at design time
The duty is on the provider and it is a design obligation, not an operational one. That matters if you bought your chatbot: the vendor must have built the disclosure in, but if you put your own name or trademark on the system you can become the provider yourself (Article 25) and inherit the duty. Check which of the two you are before assuming the vendor covered it.
When it does not apply
Two carve-outs. First, where the AI nature is obvious from the point of view of that reasonably observant person — a clearly labelled experiment, an interface that plainly announces itself. 'Our users know' is not the test; the standard is objective. Second, AI systems authorised by law to detect, prevent, investigate or prosecute criminal offences, subject to safeguards — but not where those systems are available to the public to report an offence.
What it does not require
It does not require a conformity assessment, registration or a specific banner design. It requires that the person is informed. A persistent, plainly worded statement at the start of the interaction is the common way to meet it; burying it in a privacy policy is not.
The Commission's own guidance
The Commission published guidelines on the Article 50 transparency obligations on 20 July 2026, applying from 2 August. They restate the design duty in plain terms: providers must design AI systems so users are informed when they are interacting directly with an AI. Guidelines are not binding, but they are the clearest statement of how the Commission reads the paragraph, and a supervisor will have read them too.
What to do
List every system in your organisation that talks to a person — website chat, phone bots, in-app assistants, automated email replies. For each, establish whether you are provider or deployer, confirm the disclosure is present at the start of the interaction rather than in a policy document, and record that assessment with its date. If a vendor supplies the system, ask them in writing how they meet Article 50(1).
Sources
Last verified against the primary sources: 2026-08-04
Check your own systems
The free
Article 50 check maps one system to the transparency duties in about two minutes, quoting the official text per answer. No sign-up, and the full result before any contact details.
← EU AI Act overview