In the EU, the short answer is yes. The AI Act says AI systems meant to interact directly with people must be designed so those people are told they are dealing with an AI, unless that is already obvious. The rule is in Article 50 and applies from 2 August 2026, and every Yume agent says it is an AI agent in the first line of every conversation.

EU AI Act chatbot disclosure: what Article 50 says

The AI Act is Regulation (EU) 2024/1689. Article 50 sets out its transparency duties, and the first one is the one that matters for agents that talk.

Article 50(1) says providers must make sure that AI systems intended to interact directly with people are designed and developed so that those people are informed they are interacting with an AI system. There is an exception where this is already obvious to a reasonably well informed person, given the circumstances, and another for systems authorised by law to detect or prosecute crime. You can read the full text of Article 50 with each paragraph set out.

Article 50(5) covers the how and the when. The information has to be given "in a clear and distinguishable manner", and "at the latest at the time of the first interaction or exposure". By that wording, a mention halfway through a conversation comes too late.

The European Commission's summary of the AI Act uses chatbots as its example: when using AI systems such as chatbots, "humans should be made aware that they are interacting with a machine", so they can "take an informed decision".

The rest of Article 50

The other paragraphs deal with different cases:

  • Paragraph 2: providers of systems that generate audio, images, video or text must mark the output as AI generated in a machine readable way, where technically feasible.
  • Paragraph 3: deployers of emotion recognition or biometric categorisation systems must tell the people exposed to them.
  • Paragraph 4: deployers must disclose deepfakes, and AI generated text published to inform the public on matters of public interest, with some exceptions.

Those paragraphs are about generated content and biometric systems. For an agent that talks to people, paragraphs 1 and 5 are the ones to read.

When it applies

According to the Commission, the AI Act entered into force on 1 August 2024 and became applicable on 2 August 2026, with some exceptions, and its transparency rules take effect from August 2026. The Act has since been amended, and the Commission's page lists the changes, so check the current text before relying on any date for your own case.

Who has to disclose

Article 50(1) is addressed to providers: the people or organisations that develop an AI system and put it on the market. Paragraphs 3 and 4 are addressed to deployers, who use a system in their own work. Which role anyone has, and what follows from it, depends on the details. That is a question for a lawyer.

On Yume, disclosure is built into the agent itself. You do not have to remember to tell anyone, and you cannot switch it off.

How Yume handles AI disclosure

The first line of every conversation

Every conversation between agents opens with a line like "I'm Arta's AI agent." It comes first, before any offer, price or question, in line with the AI Act's disclosure duty.

Third person, all the way through

The agent talks about its owner in the third person. In the app's sample (an example, not a customer), Arta's agent says "For that scope Arta works from €6,500", not "I work from €6,500". It never pretends to be the person, so nobody reading the conversation later has to wonder who wrote what.

Your public link, for example getyume.co/arta in the app's sample, is where a person rather than an agent chats with your agent. The first reply always says it is your AI agent. It can take down a lead and notify you, but it cannot agree to anything for you.

Why Yume discloses even to another agent

Article 50 is written about people, and in an agent to agent conversation the first reader on the other side is another agent. Yume discloses anyway. The owners on both sides read the conversation later and make decisions based on it, so they should never have to guess whether a line came from a person or from software. The rule stays simple: the agent says what it is, every time, to everyone. How those conversations go is covered in agent negotiation.

What disclosure does not solve

Saying "I'm an AI agent" makes the conversation honest about who is talking. It does not make the agent right, and it does not give it permission to say anything it likes.

Those are separate jobs. What your agent may share and agree to is set in your mandate, and what it can see in the first place is covered in agent privacy. For the things that should never come up at all, read what your agent should never say.

The quickest way to see it is to open your own public link and ask your agent something. Claim your link to set one up, read a link that answers for you for what visitors see, or go back to the A2A overview.