Rules and compliance

The EU AI Act and Voice Agents: Transparency Basics

What the EU AI Act's transparency duties mean for voice agents that answer or place calls, with a checklist for telling callers they are speaking to AI.

Call editorial team · Published · 6 min read

Voice agents are good enough now that a caller can finish a conversation without realising there was no person on the line. European lawmakers noticed. The EU Artificial Intelligence Act (Regulation (EU) 2024/1689) includes transparency rules aimed squarely at systems that interact with people, and a phone agent is a textbook example.

A person at a desk wearing a headset while a call is shown on a laptop screen
Illustrative image

This guide explains the basics in plain language: what the transparency duty says, who it falls on, what it means for inbound and outbound calls, and how to build it into your call flow. It is an overview for operations and product teams, not a legal opinion. For the wider picture on telling callers about AI, see Telling Callers They Are Talking to an AI.

The Act in one minute

The AI Act sorts AI systems by the risk they pose and attaches duties to each tier. A small number of practices are prohibited. A defined list of high-risk uses carries heavy obligations. Systems that interact directly with people, or that generate synthetic audio, image or video, carry lighter transparency obligations. Most general business uses fall outside the high-risk list.

A customer-service voice agent usually sits in that transparency tier rather than the high-risk one. That does not make it unregulated: other laws still apply, such as data protection rules and consumer law. But the AI Act's own requirement for this kind of system is mainly about honesty.

Where a voice agent fits

  • Prohibited practicesManipulative or exploitative uses are banned outright. An ordinary receptionist or support agent is not designed for these, but your prompts and scripts should never drift toward them.
  • High-risk systemsA fixed list of uses, such as certain decisions about employment, credit or essential services. Check whether your call flow makes or heavily influences such decisions.
  • Transparency dutiesThe tier most phone agents fall into: people must be told they are interacting with an AI system, and synthetic audio should be marked.
  • Everything elseOther laws still apply, including data protection, recording rules and consumer protection.

What the transparency duty says

In plain terms, the Act's transparency article asks for the following, with some exceptions:

Under the original timetable, the transparency obligations are scheduled to apply from August 2026. Timetables for parts of the Act have been under discussion at EU level, so confirm the dates that apply when you read this.

Is it ever obvious?

The Act allows an exception when it is obvious to a reasonably informed person that they are dealing with an AI. On the phone, do not lean on this. A natural voice, a human-sounding name and a friendly tone make it far from obvious. If you are unsure, disclose.

Provider or deployer: who has which duty

The Act distinguishes providers, who develop an AI system or have one developed and place it on the market under their own name, from deployers, who use a system under their authority. A business that configures a voice agent from a platform and puts it on its phone line is typically a deployer, while the platform vendor is closer to a provider. Many duties sit with the provider, but a deployer still has to use the system as intended, keep humans informed where required and not strip out the safeguards.

Provider and deployer, side by side

Provider (builds or sells the system)Deployer (puts it to use)
Typical exampleA vendor of a voice-agent platformA clinic, retailer or service desk that runs a voice agent on its line
Disclosure to peopleBuilds the system so it can tell people they are dealing with an AIMakes sure the line actually tells callers, in the right place and in the right words
Synthetic audio markingBuilds in marking where technically feasibleDoes not remove or disable it
Day-to-day controlDocumentation and designScripts, escalation, monitoring and record keeping

If you are not sure which role you hold, assume you have at least the deployer duties and ask your vendor what the platform does for you.

Inbound and outbound calls

Inbound calls are the simpler case. The caller chose to ring you, and you can open with a short statement before the agent takes over. Outbound calls need more care because the person did not choose the conversation. The disclosure should come in the first sentence or two, along with who is calling and why.

A transparent opening for an AI-handled call

  1. Name the business"Hello, this is the booking line at Harbour Physio."
  2. Say it is an AI"I'm an automated voice assistant, not a person."
  3. Say what it can do"I can take a message or a request for an appointment."
  4. Offer a person"If you'd rather speak to someone, just say so."
  5. Mention recording if relevant"This call may be recorded for quality."

These examples are hypothetical; adapt them to your own business. For outbound campaigns also keep to the separate rules on automated calls and consent, summarised in Automated Calls: A Country-by-Country Overview.

Synthetic and cloned voices

Text-to-speech voices are synthetic audio, and a cloned voice goes further by imitating a real person. Marking duties fall mainly on the provider, but as a deployer you should understand what your platform does, and you should never use a cloned voice of a real person without that person's clear permission. If your agent speaks in the voice of a known employee, tell callers it is an AI voice, not that employee.

For more on voices, see Voice Cloning for Business: Consent and Limits.

Beyond disclosure: what else to design in

Transparency is the floor, not the ceiling. A well-run voice agent also needs these:

Practical safeguards around the disclosure

  • A way to a personCallers who ask for a human should reach one, or be offered a callback or message, without a fight.
  • Limits on what it saysGuardrails that stop the agent inventing answers or giving advice outside its remit.
  • Data careCollect only what the call needs, and tell callers how recordings and transcripts are used.
  • Human oversightSomeone reviews transcripts, handles escalations and can switch the agent off.

See Guardrails for Voice AI Agents and Privacy by Design for Voice AI for the design side.

A practical readiness checklist

Before an AI voice agent takes live calls

  • Decide whether you are a provider, a deployer or both for this system
  • Write the disclosure in plain words and place it first on every call type
  • Make sure a caller can reach a person, a callback or a message
  • Confirm with your vendor how synthetic audio is marked
  • Check that data protection and recording notices are in place as well
  • Keep records of the wording used and the dates it changed
  • Review a sample of real calls each week and fix any that hid the AI
  • Re-check official sources for timetable changes before launch

Doing it in Call

In Call, an operator can list required disclosures under Admin, Telephony & AI: you give a label, the words that count as saying it, and the number of seconds by which it must have been said. Only the agent's own speech counts, not a caller asking about it. On a phone line you can also choose whether callers hear a recording and monitoring notice, and on AI calls that notice comes before the greeting. A caller can ask for a person, and AI agents can hand a call over; see Handing Calls from an AI Agent to a Person in Call. Whether those settings satisfy a particular legal duty is for you and your adviser to decide.

Frequently asked questions

Does the EU AI Act apply to me if my business is outside the EU?

It can. The Act reaches providers and deployers outside the EU when their system's output is used in the EU, for example by callers located there. Take advice if you have EU callers.

Must a voice agent always say it is an AI?

The transparency rule asks that people be informed they are interacting with an AI system unless that is obvious to a reasonably informed person. On a phone line it is rarely obvious, so the safe habit is to say so near the start of every call.

When do the transparency rules start to apply?

Under the original timetable they are scheduled to apply from August 2026, but timelines for parts of the Act have been under discussion. Check the European Commission's pages for the dates in force when you read this.

Is a text-to-speech greeting an AI system?

A fixed recorded or synthetic greeting that does not interact is different from an agent that holds a conversation. Marking duties for synthetic audio may still be relevant, so ask your vendor how it handles them.

Does disclosing the AI mean callers will hang up?

Some may, but honest disclosure builds trust, and callers who stay tend to be clearer about what they want. A friendly opening and an easy route to a person reduce drop-off.

Is the AI Act the only law that matters for voice agents?

No. Data protection, call recording, marketing and consent rules, and sector rules still apply. The AI Act adds to them rather than replacing them.

Set up a clear AI disclosure for your phone line in Call

In the demo console you can try incoming calls in a simulation.

Related guides