Does your chatbot have to say it is AI? What Article 50 has required since 2 August 2026

Since 2 August 2026, every provider of an AI system intended to interact directly with natural persons must design that system so the person knows they are communicating with an AI system. This follows from Article 50(1) of the AI Act and applies, for example, to chatbots and voice assistants. The duty only falls away when it is evident to a reasonably well-informed, observant and circumspect person, given the circumstances and context. The duty was not postponed by the Digital Omnibus and has no transition period: it has applied in full since 2 August 2026.

This page accompanies our practical overview of Article 50. Because disclosing alone is not enough, you will also find below why steering behaviour by a chatbot remains a real risk even when the AI disclosure is in place.

What does Article 50(1) require?

Article 50(1) requires the provider to design an AI system that interacts directly with people so the user knows they are communicating with an AI system. It is therefore not a separate notice given afterwards, but a design duty: the knowledge that this is AI must be built into the system itself.

The disclosure must be given at the latest at first interaction, be clear and distinguishable, and be accessible to people with disabilities. A disclosure that is effectively impossible to find does not comply.

Who must make the disclosure?

The duty sits with the provider: the party that develops the AI system or has it developed and places it on the market or puts it into service under its own name or trademark. Because it is a design duty, the disclosure belongs to the system itself, not to its incidental use.

A short, clear sentence at the start of the conversation, such as "You are talking to an AI assistant, not an employee", can be enough (guidelines, point 37). Where organisations go wrong is a notice that only sits in the terms or metadata, or an ambiguous label such as just "assistant" (guidelines, point 38). The disclosure must be clear and distinguishable, and the design must carry that knowledge from the first contact.

When is the disclosure not required?

The duty is not absolute. It falls away when it is evident to a reasonably well-informed, observant and circumspect person that they are dealing with an AI system, given the circumstances and context. There is also an exception for AI systems authorised by law to detect, prevent, investigate or prosecute criminal offences, subject to appropriate safeguards. That exception in turn falls away when the system is available for the public to report criminal offences.

Anyone relying on the obviousness exception must be able to explain why it is genuinely clear to an average user. When in doubt, the exception does not apply: the guidelines limit it to cases with almost no doubt (guidelines, point 45). For a public system that children, older people or persons with disabilities may use, you cannot rely on it (guidelines, point 44).

Disclosing alone is not enough

The most underestimated mistake is to think your concern ends with the notice that this is AI. Article 50(1) only requires that notice, but steering by a chatbot can engage other rules, such as the ban on manipulative techniques in Article 5 and consumer law. A chatbot that steers users remains a risk even when it neatly discloses that it is AI. The Dutch Data Protection Authority found that chatbots systematically gave coloured voting advice. Implicit steering is just as risky as explicit advice.

How does this relate to the rest of Article 50?

This disclosure duty for direct AI interaction sits apart from two other duties. The machine-readable marking of AI-generated content is in Article 50(2). The visible disclosure for deepfakes is in Article 50(4). A chatbot may fall under 50(1) while the content it produces falls under a different rule. Keep these duties apart and determine per case which one applies.

What should you do now?

If you have not arranged this yet, do it now: the duty has applied since 2 August 2026.

Disclosure texts for the chat, the voicebot, an email from an AI agent and a deep fake, each with its paragraph and role, are in the open library of AI disclosure texts. The full Article 50 kit adds the decision tree per paragraph and an evidence log.

Want to determine which obligation applies to your own use case? Walk through the Article 50 decision tree: per role and use case the paragraph, the steps from the guidelines and examples of good notices.