AI disclosure in phone calls, what the EU AI Act asks and how to comply

Regulation (EU) 2024/1689, the EU AI Act, has an article on transparency, Article 50. Its first paragraph says providers must ensure that AI systems intended to interact directly with people are designed and developed so that those people are informed they are interacting with an AI system.

Put simply, if an AI system talks to a person, that person has to know.

Paragraph 5 adds the how and the when. The information must be given in a clear and distinguishable way, at the latest at the time of the first interaction, and it has to be accessible. On a phone call, the first interaction is the first thing the caller hears.

It has applied since 2 August 2026 (Article 113), and because it is an EU regulation it applies directly in Spain and across the Union, without waiting for a national law.

The regulation separates two roles. The provider develops the system and places it on the market under its own name or brand. The deployer uses it under its own authority, which here is your business.

The Article 50(1) disclosure is an obligation of the provider, and with Lera agents the provider is Lera. That is why the disclosure cannot be handled by a contract clause that passes it on to you. It has to be built into the product, and it is.

Note that data protection works the other way round. There you are the controller and Lera the processor, so telling callers what you do with their data, as Article 13 of the GDPR requires, is your job. They are two different notices and it is worth keeping them apart.

You still have a small part in the AI disclosure. In Lera’s Customer Agreement you commit not to remove, hide or distort that notice, and not to present the agent as a person.

Article 50(1) has an exception. No disclosure is needed when it is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and the context of use.

On the phone, do not count on it. Today’s voices sound very natural, and someone calling a shop or a clinic expects to reach someone who works there. The fact that a caller might suspect it does not make it obvious.

The European Commission’s guidelines on Article 50, published in July 2026, read this exception strictly. They are not binding, but they are what guides the authorities that enforce it.

The rule does not set any wording. What has to be clear is that an AI system, not a person, is on the other end, and the more explicit the better.

The Commission’s guidelines add a little more. In a voice conversation, the natural way is a spoken statement at the beginning. A notice tucked away in your website’s terms is not enough, and a generic reference to “an assistant” may fall short.

The simplest place is the greeting, because it is the first thing the caller hears. In Spain the most common formula is “Hola, soy el asistente virtual de [your business]”, the Spanish for “Hi, I’m the virtual assistant for [your business]”, and it is what Lera uses by default. If you want no doubt at all, name the technology, for example “Hi, I’m the AI assistant for [your business]”.

Also say on whose behalf the agent is answering. And keep it to one short sentence, because nobody listens to a paragraph when they pick up.

The default greeting says it. If you do not write your own, your agent opens with “Hi, I’m the virtual assistant for [your business]. How can I help you?”, or its Spanish or French version.

The panel warns you if your greeting loses it. It is in Agents, inside your agent, under Welcome message. If the text you write does not say that a virtual assistant is speaking, an amber warning appears. Single prompt agents, the ones that run on one set of instructions, like the briefing you would give a new employee, also get an “Add it at the start” button that puts “Virtual assistant for [your business].” in front of your greeting. Flow agents, the ones that follow a set of steps, show the warning when you save and when you publish.

If you switch off Greet on pick-up, the agent waits for the caller to speak first, and the disclosure has to go in its first sentence. In a flow agent the panel then checks the fixed sentence of the first step, if there is one. In a single prompt agent that first sentence is improvised by the agent and there is no text to check, so the safest option is to leave the greeting on.

If someone asks whether it is a person, it says no. Your agent always carries the instruction to say naturally that it is your company’s AI assistant when someone asks. It is not a panel option and it cannot be switched off.

You can see it in Analytics. It shows in how many of your calls the agent said at the start that it is a virtual assistant, counting the calls where it got to speak.

Chat works the same way. The disclosure goes in the welcome message, and the panel warns you if yours does not say a virtual assistant is writing.

What Lera does not do is stop you publishing an agent without the disclosure, or insert a sentence of its own into your calls. How to say it is your choice.

The disclosure is not only for the calls your agent receives. When it calls, to remind someone of an appointment or update them on an order, its first sentence has to say who is calling, on behalf of which business, and that it is a virtual assistant.

The easiest way is to put it all in the same sentence as the reason for the call, and to ask for exactly that in the instructions the agent opens the call with. “Hi, I’m the virtual assistant for [your business], I’m calling to remind you of your appointment tomorrow at ten.”

That also applies when it calls professionals, such as outside technicians or suppliers. Working with you does not make it obvious that the caller is an AI system.

Is it enough to put it in my website’s terms? No. The regulation requires the disclosure to reach the person talking to the agent clearly, at the latest at the first interaction. On a call that is the greeting, not a legal text almost nobody reads.

Do I have to repeat it during the call? The rule asks for it at the start at the latest. If someone asks halfway through whether they are talking to a person, your agent always tells them it is not.

Can I say “virtual assistant” instead of “artificial intelligence”? It is the most common formula in Spain and the one Lera uses by default. The rule does not require specific words, but the clearer the better, and naming AI leaves no doubt.

What happens if my greeting does not say it? The panel warns you, but it does not stop you publishing. Bear in mind that the regulation requires it and that in the Customer Agreement you commit not to remove the disclosure.

Does this replace the data protection notice? No. Saying that an AI is speaking is one thing, and saying who processes the caller’s data, why and how to exercise their rights is another. The second one is required by the GDPR and is yours.

Is this legal advice? No. It is a practical explanation of the rule as it applies today. If your case has something special, such as calls about health or with vulnerable people, review it with your legal adviser.