Does your AI phone assistant have to tell callers it's AI? (EU, 2026)

If you run a business phone line answered by an AI assistant — a receptionist that books appointments, answers common questions, or takes messages — you've probably asked the obvious question: do we have to tell callers it's AI? It's the first thing careful owners raise, and it's the exact point one of our clients wanted nailed down in writing before signing. Here's the straight answer, what the rule actually says, and how we implement it.

Do you have to tell callers they're talking to AI?

In the EU, yes. The EU AI Act sets transparency obligations for AI systems that interact directly with people. In plain terms: if a system talks to a person, the person should be able to know they're dealing with AI rather than a human — unless that's already obvious from the context. A phone line where an AI books your car service is not obvious to a first-time caller, so it needs a brief heads-up.

The important nuance: the law asks you to inform the caller, not to interrogate them or read a paragraph of legalese. There's no requirement for a recorded disclaimer, a keypad menu, or an automatic hand-off to a human. A clear, natural statement that the caller is speaking with an AI assistant satisfies the intent.

When does the rule apply?

The EU AI Act entered into force on 1 August 2024, and its obligations switch on in stages. The transparency rules that cover AI you talk to — chatbots and voice assistants included — apply from 2 August 2026. If your AI phone line serves callers in the EU, that's the date to plan around. Even outside the EU, disclosure is fast becoming the norm and a trust signal, so there's little reason to wait.

How do you disclose it without hurting the call?

The whole thing is one sentence, spoken naturally at the very start, before the caller has to explain why they rang. It names the business, says it's the AI assistant, and moves straight into helping. Callers barely register it as a compliance line — it just sounds like a modern, well-run front desk.

“Hi, you've reached the AI assistant for Miller's Auto Service. How can I help?”

That's it. No second sentence, no "this call may be recorded for AI training," no "press 1 to continue." The disclosure is present, natural, and over in two seconds — which is exactly what keeps the experience good while meeting the requirement.

How we build compliant disclosure into a voice assistant

We treat the opening line as a fixed, non-skippable part of the call flow, not something the model can drift away from. A few engineering details make it reliable:

• The AI identifier is in the greeting itself, so it's said on every call regardless of how the conversation goes.
• It's templated per business and per language, so the wording is natural in the caller's language, not a translated afterthought.
• It's kept short and warm on purpose — the goal is a good first impression that also happens to be compliant.
• If a caller asks to speak to a human, the assistant hands off cleanly — disclosure and escalation are separate features, and both are there.
• The greeting is logged, so there's a record that every call opened with the disclosure.

If you're weighing an AI phone assistant and want it done right — compliant, natural, and genuinely useful from the first call — that's the kind of thing we build and run end to end. Tell us about your setup and we'll scope it with you.