Skip to main content
FAQ: Frequently Asked Questions

AI Act: What You Need to Do to Ensure Your Agents Are in Compliance

A practical checklist for configuring your agents in compliance with regulations—no surprises.

Written By Product

Last updated 2 months ago

As of August 2, 2026, Article 50 of EU Regulation 2024/1689 (AI Act) will be fully in effect: anyone speaking with your chatbot must understand right away that they are speaking with an AI. This isn’t a detail to handle “if it comes up”—it must be built into the chatbot from the very beginning. Here’s how to stay compliant without having to think about it every time.


The basic rule: disclose it; don’t wait to be asked

The obligation doesn’t end with honestly answering if the contact asks, “Are you an AI?” It goes further: the person must understand this from the very first exchanges of the conversation, unless it’s already evident from the context (e.g., a chat widget that clearly identifies itself as a bot).

In practice, this means structuring the agent’s opening so that its artificial nature comes across naturally—in the greeting, in the way it introduces itself—without the user having to ask.

How to write it in the prompt

  • Include it in the opening, not just in the objection handling section. It’s not enough to prepare a ready-made response for when they ask: the statement should be integrated into the initial greeting or the very first few lines.

  • It doesn’t need to sound legalistic. A natural line (“I’m the virtual assistant for [company]; I’m calling you to…”) is enough and fits into the script better than you might think.

  • If they ask anyway, always confirm it. No exceptions: if a contact explicitly asks, “Are you an AI/bot?”, the agent must answer yes.

Can I hide it?

Out of fear of negative reactions, some have asked us not to disclose this. Our policy here is clear: we absolutely cannot accommodate this request.


Call Recording

The same logic applies as with AI disclosure: if the agent is recording the conversation, the contact must know this from the start, before the recording begins—not find out after the fact or have to ask for it. It must be stated at the beginning, along with (or immediately after) the introduction as an AI.


Transferring to a Human Agent

There’s no need to transfer the call immediately every time the contact asks to speak with a person. It is sufficient for the agent to gather the necessary information and open a ticket so a colleague can call them back—provided that this human support channel actually exists and the request is genuinely handled, not just verbally promised by the agent.


Penalties

This is not a formality to be put off: violations of the transparency obligations under Article 50 can cost up to 3% of the company’s annual global revenue or 15 million euros, whichever amount is higher.


Quick Checklist

The starting point varies depending on the type of agent:

  • Outbound: The disclosure must be made at the beginning of the call, even before explaining the reason for the call

  • Inbound: It must be included in the greeting the agent uses when answering, as soon as the contact connects

  • The agent identifies themselves as an AI/virtual assistant right from the start, not just upon request

  • If explicitly requested, the agent always confirms that they are an AI

  • If the agent is recording the call, they must state this at the beginning, before the recording starts

  • There are no instructions in the prompt telling the agent to deny its nature

  • There is a real channel for transferring the call to a human when requested