Incident Report

Air Canada Chatbot Refund Promise

An airline chatbot fabricated a bereavement refund policy, binding the company to a promise no human agent would have made.

Summary

Canadian tribunal ruled the airline liable for the chatbot's fabricated policy.

Timeline

Passenger queries airline about bereavement fare refund policy

Chatbot provides fabricated refund policy with specific deadline and process

Passenger submits refund claim based on chatbot's instructions

Airline denies refund, claiming chatbot's policy was inaccurate

Passenger files complaint with Canadian tribunal

Tribunal rules airline liable for chatbot's representations

Root Cause

The chatbot was allowed to generate policy information without guardrails, leading to fabricated refund procedures that contradicted actual airline policy.

Contributory Factors

  • No human oversight of chatbot responses
  • Chatbot not trained on verified policy documents
  • No disclaimer about chatbot accuracy
  • Airline attempted to disclaim liability for chatbot output

Sources

Analysis

This incident demonstrates a critical failure in AI deployment: allowing an automated system to make authoritative policy statements without verification mechanisms.

The chatbot’s fabricated refund policy was plausible enough that a reasonable person would accept it as accurate. The airline’s attempt to disclaim responsibility failed because the tribunal determined that presenting a chatbot as an official communication channel created reasonable reliance.

Key Takeaways

  1. Chatbots must be constrained to verified information — generative responses about policies, prices, or legal obligations create liability.

  2. Disclaimer disclaimers are insufficient — telling users “this may not be accurate” does not eliminate the organization’s responsibility for what the system actually says.

  3. Human escalation paths must be clear — if a chatbot cannot reliably answer a question, it should route to a human agent rather than fabricate an answer.