Skip to main content

Customer Support Chatbot

This page covers support chat interfaces where users interact with an AI assistant.

Why this is usually transparency-triggered​

When users cannot reasonably tell they are interacting with AI, disclosure is required. The core obligation here is not usually high-risk control packs, but clear and reliable transparency implementation.

Whose duty is it? (Art 50(1))

The obligation falls on the provider: design the system so users know they are dealing with AI. The duty does not apply where AI interaction is obvious to a reasonably well-informed, observant person in the circumstances. As deployer, still verify the disclosure works in your product context.

Required obligations​

WHEN​

  • Users chat with automated support in product, website, or support channels.
  • The interaction could reasonably be mistaken for human-only support.

MUST​

  • Add clear AI-interaction disclosure at the point of interaction.
  • Keep disclosure understandable, visible, and consistent across supported locales.
  • Validate disclosure behavior during release checks.

OUTPUT artifacts​

  • UI specification/screenshots showing disclosure placement and exact copy.
  • Localization verification records for supported languages.
  • Release checklist evidence confirming disclosure validation.
Real-world illustration

Moffatt v. Air Canada (Canada, 2024): a tribunal held the airline liable for wrong information its support chatbot gave a customer, rejecting the argument that the chatbot was a separate legal entity. Decided under consumer/contract law, not the AI Act — but it shows deployers own their chatbot's behavior, alongside the Art 50(1) duty to tell users they are dealing with AI. (BBC)