Do you have to tell customers they are talking to AI?

Do you have to tell customers they are talking to AI?

In the EU, yes: from August 2, 2026, AI Act Article 50 requires chat and voice agents to say they are AI unless it is obvious. The US has no single federal rule, but Maine and California have bot disclosure laws.

This page goes deeper on one question our AI customer service agent service page answers in a few lines. We build every support agent to say it is an AI in its first message, so for our clients the answer is settled in practice. The detail below is for owners who want to know which rule applies where, what the official texts actually say, and what "clear and conspicuous" looks like in a chat window or on a phone call.

We read each text on the legislature's or regulator's own site on October 30, 2026, or, where the statute page was blocked, the legislature's own committee analysis of it. This is a plain summary, not legal advice. Laws in this area change every session, so check your own markets with counsel.

Which laws require AI or bot disclosure, and what do they say?

The EU AI Act requires disclosure for AI that talks to people. Maine bans chatbots that could seem human to a reasonable consumer unless clearly labelled. California bans deceptive bots that push a sale. Colorado adds a duty in 2027.

Two things stand out from the official texts. First, the EU rule and the Maine rule apply to ordinary customer service: answering an order question is "trade and commerce" in Maine, and a support chatbot plainly "interacts directly" with people under the EU Act. Second, California's 2018 law is narrower than its reputation. It targets bots used to deceive people into a sale or a vote, and disclosure is its safe harbour, so a support agent that says it is a bot is outside it.

California has tried to widen its rule. Assembly Bill 410, introduced in 2025, would have amended sections 17940 to 17942 and rewritten 17943, according to the Assembly committee's analysis. We could not confirm from the official bill page on October 30, 2026 whether it became law, so treat it as pending and check its status.

Colorado is next. HB 26-1263, signed on May 29, 2026, requires operators of conversational AI services open to the general public to provide a disclosure that the service is artificial intelligence, among other duties; the bill summary says the operator requirements take effect January 1, 2027. Our guide to US state AI laws for businesses covers Colorado and Texas in detail.

Does the EU rule apply if your business is outside the EU?

It can. Article 2(1)(c) covers providers and deployers in third countries "where the output produced by the AI system is used in the Union". If EU residents chat with your support agent, plan for the Article 50 disclosure from August 2, 2026.

Article 50(1) places the design duty on the provider: the agent must be "designed and developed in such a way" that people are told. When you use Intercom, Zendesk or a model API, the vendor is usually the provider and you are the deployer, so your job is to keep the disclosure switched on and visible. When we build a custom agent for you, the opening message is part of the build. Our page on whether the EU AI Act applies to US companies covers the provider and deployer roles in more detail.

What counts as clear and conspicuous disclosure in a chat or call?

Say it in the agent's first message or first spoken sentence, in plain words such as "I'm an AI assistant", where the customer will see or hear it before they share anything. A note in your terms of service does not meet that standard.

None of the texts we read gives a script, so we design to the strictest wording we read: the EU's "at the latest at the time of the first interaction". In practice that means:

• Chat and WhatsApp: the first message says the customer is talking to an AI assistant and how to reach a person. The widget header can repeat it ("AI assistant").
• Email: the signature names the AI assistant, and the first reply in a thread says it.
• Phone: the greeting says it in the first sentence, before any question. For outbound calls, consent rules also apply; see our guide to whether AI calling is legal.
• When asked: if a customer asks "am I talking to a real person?", the agent answers truthfully, every time, with no clever deflection.
• Never a human name with no label: a friendly name is fine ("Ava, our AI assistant"), a name that suggests a person is not.

What happens if you do not disclose?

In the EU, breaching Article 50 can bring fines up to EUR 15 million or 3% of turnover. In Maine it is an unfair trade practice. Everywhere, an agent that pretends to be human invites deception claims and damages trust when customers find out.

Regulators already treat claims about AI like any other business claim. In February 2025 the US Federal Trade Commission finalized an order against DoNotPay, which it said "did not test whether its 'AI lawyer' operated to the level of a human lawyer", and required it to pay $193,000. That case was about overclaiming what AI can do, not about disclosure, but it shows how US deception law reaches AI products. A related question, who pays when the agent gets something wrong, is answered on our page about whether a company is liable for what its chatbot says.

Does disclosure hurt how customers use the agent?

We have no study of our own to quote, and we do not rely on vendor surveys. What disclosure does in practice is set expectations: customers know they can ask for a person, which makes the handoff feel normal instead of a failure.

The bigger risk is the opposite: a customer who believed they were talking to a person and later learns otherwise. That is the situation these laws are written to prevent. Disclosure works well paired with a clear way out. Our money page sets out how we design the handoff from AI to a person, and our page on whether an AI agent will replace staff covers how the two work together.

More questions about AI disclosure

Is there a federal law in the US that requires AI disclosure?

We found no general federal statute that requires every business chatbot to say it is an AI. The FTC can act against deceptive practices, which can include misleading people about whether they are talking to a person or about what an AI can do. Phone calls are different again: AI voice calls fall under federal calling and consent rules, which our AI calling guide covers.

What is the California B.O.T. Act?

It is the common name for SB 1001, passed in 2018 and codified at Business and Professions Code sections 17940 to 17943. It makes it unlawful to use a bot online to mislead people about its artificial identity in order to knowingly deceive them into a purchase or sale, or to influence a vote. Disclosing that it is a bot, clearly and conspicuously, is a defence.

What does Maine's chatbot disclosure law require?

Maine's law, 10 M.R.S. section 1500-DD, says a person may not use an AI chatbot or other computer technology in trade and commerce in a way that may mislead or deceive a reasonable consumer into believing they are engaging with a human, unless the consumer gets clear and conspicuous notice that they are not. A breach is treated as a violation of the Maine Unfair Trade Practices Act.

Do UK businesses have to disclose AI chatbots?

We did not find a UK statute that specifically requires chatbot disclosure, and we have not reviewed UK consumer law for this page. If you also serve EU customers, the EU AI Act rule applies to those conversations from August 2, 2026, wherever your company is based. A UK business should check general consumer protection rules on misleading practices with its own adviser.

Is it enough to mention AI in the privacy policy?

No. The EU text asks for disclosure "at the latest at the time of the first interaction", and Maine and California both ask for clear and conspicuous notice to the person the bot is talking to. A policy page most customers never open does not meet that. Put it in the first message or the first spoken sentence, and keep the privacy policy link close by.

Can an AI agent use a human-sounding name?

It can use a name, as long as it is labelled as an AI in the same breath, for example "Ava, the AI assistant for our clinic". The problem is a name with no label, especially on a voice call where the voice sounds human. Under Maine's rule the test is whether a reasonable consumer could be misled into thinking they are dealing with a person.

Do voice agents need to disclose too?

Yes. The EU rule covers any AI system that interacts directly with people, which includes voice. Maine's definition of an AI chatbot covers text or voice communication. California's Public Utilities Code also requires automatic dialing-announcing devices to tell the person called if a prerecorded message uses an artificial voice, according to the Assembly's analysis. See our AI receptionist service for how we script voice greetings.

Who should check the rules for my business?

Your own lawyer, for the markets you sell into. We set up the disclosure, the handoff path and the logging during a build, and we can show you exactly what the agent says in each channel. Interpreting how a law applies to your company is legal advice, which we do not give. A free strategy call is a good place to list what needs checking.

Sources and further reading

• European Commission AI Act Service Desk: Article 50, transparency obligations, checked October 30, 2026
• European Commission AI Act Service Desk: Article 2, scope, checked October 30, 2026
• European Commission AI Act Service Desk: Article 113, entry into force and application, checked October 30, 2026
• European Commission AI Act Service Desk: Article 99, penalties, checked October 30, 2026
• Maine Revised Statutes, Title 10, section 1500-DD: required disclosure of use of artificial intelligence chatbot, checked October 30, 2026
• Colorado General Assembly: HB26-1263, Conversational Artificial Intelligence Service Operator Requirements, checked October 30, 2026
• California Assembly Committee on Privacy and Consumer Protection: analysis of AB 410 (2025), including existing law at Bus. and Prof. Code sections 17940 and 17941, checked October 30, 2026
• FTC: FTC finalizes order with DoNotPay, February 11, 2025, checked October 30, 2026