Telling people they are talking to a machine used to be good manners. In the European Union it is now a rule.
What changed on 2 August 2026
Article 50 of the EU AI Act began to apply on 2 August 2026. For chatbots and voice agents, it expects that systems which talk directly to people are designed so those people are told they are dealing with AI, unless that is obvious to a reasonably well-informed, observant person. The information must be clear, given at the latest at the first interaction, and accessible.
In July 2026 the European Commission approved the content of its guidelines on Article 50. They name phone and customer-support agents as in scope. For a voice call, the example they give is an explicit spoken statement at the start, with a reminder in longer calls or after a handoff. Putting it only in the terms and conditions, behind a link, or in a generic "we use AI" notice is not enough. The guidelines had not been formally adopted at the time of writing.
Fines for breaching Article 50 can reach EUR 15 million or 3% of worldwide turnover, whichever is higher, and the lower of the two for small businesses. National market surveillance authorities do most of the enforcing.
A common mix-up
The Digital Omnibus on AI, in force from 27 July 2026, delayed the high-risk rules to December 2027 and August 2028. It did not delay Article 50's disclosure duty. Its only change for Article 50 is a grace period to 2 December 2026 for machine-readable marking of AI-generated content, and only for generative systems already on the market. A watermark is also not a substitute for telling the caller: the disclosure needs a statement the person can perceive.
Elsewhere
UK: no AI-specific disclosure law. The Competition and Markets Authority's March 2026 guidance says consumer law applies to AI agents and suggests labelling so customers are not misled about who they deal with.
India: the IT Rules amendment in force from 20 February 2026 requires labels on synthetic audio and visual content. It does not require a chat or voice agent to announce itself.
US: no federal chatbot-disclosure law. The FCC ruled in 2024 that AI-generated voices on calls fall under the TCPA consent rules. State laws vary.
Six habits for a voice or chat agent
Say it first, in the first seconds of a call or the first chat turn. For example: "You're speaking with [Company]'s virtual assistant, which is an AI. I can pass you to a colleague at any time."
Say it out loud. A spoken line, not a clause in the terms.
Always answer honestly if a caller asks whether they are speaking to a person.
Offer a person. In Gartner's 2026 survey, 87% of customers said a human option is essential.
Build it in once at platform level, and keep a record of the wording and when it played.
Treat voices with care. Cloning a real person's voice raises labelling and consent questions.
Read the full report
Our industry report Say it is an AI: the new rules on telling people who they are talking to has the timeline, the myth-versus-reality section and every source.
View the full report on our Resources page →
This article is general information based on public sources checked on 23 October 2026. It is not legal advice. Rules in this area are changing, so take advice before relying on it.
