Air Canada's chatbot ruling shows up again, this time on a rail carrier.
A Canadian provincial small claims tribunal ruled in March 2026 that a rail carrier's consumer chatbot had promised a fare refund under conditions the carrier's public policy did not permit. The tribunal cited Moffatt v. Air Canada as squarely on point: a company is responsible for the content its AI agent produces to a customer, regardless of whether the vendor, the training data, or a hallucination is the underlying cause. The carrier argued that the bot's output was an isolated error. The tribunal was unmoved.
The award was small. The precedent is not. Consumer counsel in three markets have already flagged the ruling in briefs against airline, hotel, and telecom chatbots. The template is portable across industries and across jurisdictions that recognize agency principles.
CanLII decision reference →Every chatbot in customer channel is now issuing enforceable statements on behalf of the company. The vendor cannot indemnify the customer's public promise. The transcript is the contract.
The interesting operational question is what the enterprise wants to know before the tribunal. It wants a per interaction record that shows what the bot said, what policy was active, and what the bot should have said under that policy. If the two disagree, that is a coaching signal, not a settlement. If there is no record, the transcript wins.
- transcript.exact: token by token, with model version and system prompt at that moment
- policy.snapshot: refund policy, promotion policy, disclaimer language active at that moment
- policy.applied: whether the response was allowed, transformed, or should have been blocked
- reviewer.override: whether a human accepted or corrected the answer
- export: single query trail for tribunal, carrier, and customer service leadership
Chatbot in customer channel today?
Thirty minutes on the per interaction record you would need under Moffatt style claims.
Book a working session →