The Build–Test–Revise Loop /news

Real case· 2024

Air Canada's chatbot invented a refund policy — and the airline had to pay

What happened

Booking a flight to his grandmother's funeral, Jake Moffatt asked Air Canada's website chatbot about bereavement fares. It told him he could claim the discount retroactively, within 90 days of booking. That was false — the real policy doesn't allow retroactive bereavement claims. Air Canada argued the chatbot was “a separate legal entity responsible for its own actions.” The British Columbia tribunal disagreed (Moffatt v. Air Canada, 2024 BCCRT 149), found negligent misrepresentation, and ordered Air Canada to pay $812. The output read as authoritative and was wrong — because nobody tested it against the actual policy. That is exactly the failure your build loop is designed to catch: AI output looks right until you run it against the real world.

What AI property explains this outcome? What would you do differently if you were the designer?

Read the source ↗