How to tell your clients they will be talking to your digital twin

It says people must know, clearly, at the latest at the first interaction. That is Article 50 of Regulation (EU) 2024/1689, the EU AI Act, and it has applied since 2 August 2026.

The article splits the duty between two roles. The provider, who builds the system, must design it so that people know they are talking to an AI. The deployer, who uses it with clients, has duties of its own when the system generates images, audio or video resembling a real person. If you have a digital twin, Legathum is the provider and you are the one deploying it with your clients.

The text includes an exception: you need not say so when it is obvious to a reasonably well-informed person that they are talking to an AI. With a twin that has your face and your voice, that exception does not help. Precisely because it looks like you, you have to say so.

The AI Act defines a deep fake as image, audio or video content generated or manipulated by AI that resembles real people and could make someone believe it is genuine. A digital twin speaking with your face and voice fits that description if someone does not know what it is.

For that case, Article 50(4) asks whoever deploys the system to disclose that the content has been artificially generated. In practice, you, the professional, have to make sure your clients know, rather than relying only on the system to tell them.

The good news is that doing it is simple and, done well, it builds trust rather than eroding it.

At three moments, and the first one is before the conversation. This table shows where the notice goes and who gives it.

Order matters. If a client finds out halfway through a meeting that they were talking to an AI, the relationship starts badly even if the law was followed.

In Legathum, the twin says it is an AI in its first turn, and its name in the meeting always ends in "Clon IA". The house twin, for example, joins Google Meet as "Deibson Clon IA".

Both follow the same rule in Article 50: the information must arrive clearly and distinguishably, at the latest at the first interaction. The name is visible before anyone speaks; the opening line confirms it as soon as the conversation begins.

And if someone asks halfway through whether they are talking to an AI, the answer must always be yes.

A few short lines are enough. Here are examples you can adapt to your own tone; replace {name} with yours and {service} with what you do.

On your website or booking page: "Part of my client work is handled by my digital twin, an artificial intelligence that speaks with my face and my voice. It will always tell you it is an AI, and if you would rather speak to me, just ask."

In the booking confirmation: "Your meeting about {service} will be with {name}'s digital twin, an artificial intelligence. If you would rather meet {name}, reply to this message."

In your email signature: "{name} · Part of my client work is handled by my digital twin, an AI."

On a video you publish: a label visible throughout, for example "AI-generated video".

Notice what they have in common: they say what it is, whose it is and how to reach a person. They do not apologise and they do not hide the notice in small print.

The most serious is hiding the notice. A grey line at the foot of the page, or a notice that only appears if the client asks, is not clear and distinguishable information.

Other common mistakes: giving the twin a name that does not make clear it is an AI; calling it "an assistant" and nothing more when it has your face and your voice; assuming the client knows because they spoke to it once before; and using videos of the twin in advertising without a label.

It is also worth not overdoing it. You do not need to repeat the notice in every sentence or turn the meeting into a technical briefing. A clear name, one line at the start and an honest answer if someone asks are enough.

Not if the notice is done well. What bothers clients is finding out late, not knowing from the start.

A clear notice changes the client's question. Instead of "who am I talking to?", the question becomes "is what it is telling me useful?". That is where what sits behind the twin matters: a brain holding your decision rules, written, reviewed and approved by you in your Book of Criteria, each one with its source.

Always offering the option to speak to you helps too. For many clients, knowing they can ask is enough for them not to need to.

The EU AI Act applies to providers and deployers outside the EU when the system's output is used in the EU. For conversations with clients in the United Kingdom, UK rules apply, and they are not the same. The practical advice does not change: tell people clearly, early and honestly that they are talking to an AI.

Article 50 has applied since 2 August 2026. Regulation (EU) 2026/1744, known as the digital omnibus on AI, gave systems already on the market before 2 August 2026 until 2 December 2026 to meet Article 50(2), which covers machine-readable marking of synthetic content.

That deadline concerns providers and technical marking. The duty to make sure your client knows they are talking to an AI has no extension.

• Today: the twin says it is an AI in its first turn and its name ends in "Clon IA".
• Today: only real people are cloned, with consent by email and a video reading a sentence, manual review before the twin speaks and revocation with one click.
• In development: each client's twin serving on video-call platforms other than Google Meet.