Air Canada Held Liable for Chatbot's Misstated Bereavement Refund Policy
Air Canada's website chatbot misstated the airline's bereavement fare policy as retroactively claimable, a customer booked on that basis, and a BC tribunal ordered the airline to pay $812.02 CAD for negligent misrepresentation.
What happened
In November 2022, Air Canada's website chatbot told customer Jake Moffatt he could book full-fare flights and claim a retroactive bereavement-fare discount within 90 days. Air Canada did have a bereavement policy, but the real policy expressly excluded retroactive claims after travel — the chatbot misstated it. Moffatt booked his flights relying on the chatbot's statement. When he later attempted to claim the discount, Air Canada refused it. Moffatt took the airline to the BC Civil Resolution Tribunal, which found negligent misrepresentation in February 2024 (2024 BCCRT 149) and ordered Air Canada to pay $812.02 CAD in total: $650.88 in damages, $36.14 in pre-judgment interest, and $125 in tribunal fees. The tribunal rejected the airline's argument that the chatbot was a separate legal entity responsible for its own statements.
What the agent did
The chatbot answered a customer's policy question with a misstatement of the airline's own bereavement fare policy, presenting the discount as retroactively claimable within 90 days when the actual policy excluded retroactive claims. The chatbot executed no transaction and committed no funds; the harm arose because its unreviewed answer was authoritative enough that the customer's reliance on it later bound the airline in law.
The irreversible effect
Tribunal judgment ordering Air Canada to pay $812.02 CAD ($650.88 in damages plus $36.14 interest and $125 in fees); legal precedent establishing that a company is responsible for information provided by its chatbot, and that customer reliance on an unreviewed automated statement can bind the organization.
Root cause
The chatbot's statements about fare policy were published to customers with no review and no check against the airline's actual policy, and the airline treated the bot's output as authoritative enough to induce reliance while disclaiming responsibility for it. The tribunal held it could not have it both ways. The incident also illustrates the adjacent risk: nothing in the deployment separated answering questions from committing the company — had the same assistant been wired to effect fare or refund decisions, no deny-by-default control, approval gate, or argument-level limit stood between a fabricated policy basis and an executed commitment.
How a maker-checker control would have refused it
This was a speech-only incident: the chatbot answered a question and executed nothing, so no proxy-level refusal fires on the answer itself, and MakerChecker does not claim to make answers accurate. Had the assistant been wired to commit fare decisions — issuing the refund or booking the discounted fare directly — MakerChecker would refuse the commitment: (1) skill_not_granted — a support-answer role holds no refund-commit grant of any kind, so deny-by-default blocks any commit attempt before an obligation attaches; (2) high_risk_requires_gate — an open-amount or non-standard-basis refund commitment is categorically refused on the proxy and must run through a governed flow with a preceding approval gate decided by a named officer who is not the requester.
Runnable reproduction
This incident ships as a runnable scenario in the open-source repository. Point the enforcement engine at the policy and watch the action get refused, with the refusal written to a signed audit record.
examples/air-canada-chatbot-bereavement-refund-binding
Accuracy and corrections
This entry describes a publicly reported incident and is compiled from the primary sources listed above. Where an account is a legal allegation rather than an established finding, the entry labels it as such. Summaries can still contain errors. If you can document a correction, email hello@makerchecker.ai and we will review and correct it, with the change noted, within 14 days.
See it for yourself
Reading is one thing. Watch it block an agent.
One command starts the demo: an agent stopped from signing off its own work, and the signed evidence file an inspector can check for themselves.
Designed against the rules your auditors already enforce.