Intercom and the EU AI Act: does your chat widget tell people it's AI?
Intercom's Fin is one of the most widely deployed AI agents in customer support, and it usually answers first, before any human teammate is involved. That makes the very first message in your Messenger the moment Article 50(1) cares about. If a visitor in the EU can start chatting without being told they're talking to AI, anyone who opens the widget can see it.
The rule itself is short. Article 50(1) of the EU AI Act requires AI systems that interact directly with people to be designed so users know they're dealing with AI, no later than the first interaction, in a clear and distinguishable way. The exception for cases where it's "obvious" is narrow: a natural-language customer-service bot doesn't qualify just because it has a robot icon. A line in your terms of service doesn't satisfy it either. And it's easy to enforce, because a regulator can open your site and start a chat.
Where the disclosure lives in Intercom
- The Messenger launcher and home screen, the entry point a visitor sees before any conversation starts
- Fin's identity block: its name, avatar, and the "AI Agent" descriptor shown in the conversation header
- Fin's opening message, the first thing sent when a conversation begins
- Handover moments, where Fin passes the thread to a human teammate and the participant changes
Your Intercom disclosure checklist
- Open your site in a private window as an EU visitor and start a chat: is it explicit you're talking to AI before or in the very first exchange?
- Check Fin's identity settings, keep the AI descriptor visible and don't rename Fin to a human-sounding persona without an AI label
- Put an explicit line in the opening message (e.g. "I'm Fin, an AI agent, I can bring in a human anytime") rather than relying on an icon
- Verify the disclosure survives on mobile and in the compact Messenger view, not just desktop
- Check the human-handover message makes the participant change clear in both directions
- Screenshot the first-interaction state and save your Messenger/Fin settings export with a date
Watch out for
The classic Intercom failure mode is cosmetic humanization: Fin renamed to a teammate-style persona with a human avatar, answering instantly with natural language. Each choice is reasonable on its own. Together they make the AI nature non-obvious, and the 'obvious from context' exception stops applying. The second one is drift, Fin ships new capabilities and settings frequently, and a Messenger update or workspace migration can quietly change what visitors see first.
Common questions
Fin already has an AI label in the header, is that enough?
A persistent label helps, but the safer pattern is layered: keep the header descriptor and add an explicit statement in the first message. Regulators have signaled that disclosure must be clear and distinguishable at first interaction, and a small header tag alone is a judgment call you don't need to gamble on.
We only use Fin for some conversations, do we still need disclosure?
Yes, wherever AI is the one interacting. Mixed routing raises the bar: if visitors can't tell which conversations are AI-led, disclose at the start of any thread Fin participates in and mark handovers clearly.