AI voice agent disclosure under Article 50: what to say and when

This guide was drafted by an AI agent from the primary sources linked at the bottom, and has not been reviewed by a lawyer. It is general information, not legal advice. The original is at https://mdodeja.github.io/eu-ai-act-article-50-guide/guides/ai-voice-agent-disclosure.html.

Short answer. A voice agent that talks to people is covered the same way a chat widget is. Say it is an AI before the conversation starts, in the caller's language, and repeat it when a call is handed over or transferred. A notice on a website the caller never sees does not count.

Does Article 50 cover voice agents?

Article 50(1) of the EU AI Act covers AI systems "intended to interact directly with natural persons". It does not limit that to text. An AI that answers the phone, books appointments by voice, or calls customers is interacting directly with people, so the same rule applies as for a chat widget: people must be told they are dealing with an AI system, unless that is obvious to a reasonably well-informed, observant and circumspect person in the circumstances.

Article 50(5) adds that the information must be clear and distinguishable and given at the latest at the first interaction. On a phone call there is no header or widget to carry the notice, so in practice that means the opening words of the call.

The duty to design the system this way sits with the provider, meaning whoever built the voice agent or sells it under their own name. If you set up a vendor's voice product for a client, the vendor is usually the provider. You still control the greeting script in most products, so it is worth checking it yourself.

When to say it

Inbound calls: in the greeting, before the caller is asked anything.
Outbound calls: in the first sentence, before the reason for the call. Someone who picks up the phone has not chosen to talk to a bot, so leaving the disclosure until later is harder to defend.
Voicemail left by an AI: at the start of the message.
Transfers: when a person takes over, say so. When the call goes back to the AI, say that too.

Wording you can adapt

Inbound:

"Hi, you're speaking with [Brand]'s AI assistant. I can help with most questions, or say 'person' at any time to reach the team."

Outbound:

"Hello, this is an automated AI assistant calling for [Brand] about your appointment on [date]."

Handover:

"I'm transferring you to [name], a member of the team." Then the person introduces themselves.

Say it in the language the caller is using. If the agent switches languages mid-call, the disclosure should be available in that language too.

Where voice setups fall short

The disclosure is only on the website or in the terms. A caller never sees those.
A human-sounding name with no AI mention. "Hi, I'm Sarah from [Brand]" on its own leaves most callers assuming a person.
The disclosure comes after a question. If the caller has already answered something, the first interaction has passed.
Cloned or synthetic voices of real people. That raises separate questions under Article 50(4) on deepfakes and under other laws. Take it to a lawyer.

Other rules to check

Article 50 is only the AI part. Recording calls, using callers' data, and making automated outbound calls are covered by other laws, including data protection and national telemarketing rules, and those vary by country. This page does not cover them.