Air Canada Chatbot
The chatbot made up a refund policy. The tribunal made the airline pay it.
What They Said
Air Canada deployed a customer service chatbot on its website in 2022 as part of a broader push to deflect call volume from human agents. The chatbot was positioned as a way for travelers to get faster answers on baggage rules, fare classes, and travel policies without waiting on hold.
When the chatbot malfunctioned in court filings, Air Canada offered a defense that became infamous: the chatbot was, the airline argued, “a separate legal entity that is responsible for its own actions.” The website disclaimers, in the company’s view, should have made customers verify any answer the bot produced.
What Actually Happened
In November 2022, Jake Moffatt’s grandmother died. He went to Air Canada’s website to book a flight to the funeral and asked the chatbot about bereavement fares. The chatbot told him he could book full price and apply for a bereavement discount within 90 days of travel. He did exactly that, kept the screenshot, and submitted his refund claim.
Air Canada refused. The actual policy required bereavement discounts to be applied before travel, not after. The chatbot had hallucinated a refund window that did not exist in any Air Canada document. When Moffatt produced the chatbot transcript, the airline’s response was that he should have clicked through to the official policy page linked elsewhere on the site.
Moffatt took the case to the British Columbia Civil Resolution Tribunal. In February 2024, tribunal member Christopher Rivers ruled in his favor, awarding C$650.88 plus fees. The decision rejected Air Canada’s “separate legal entity” argument outright: “Air Canada is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.” The ruling was reported globally and became the first widely cited precedent that companies are legally bound by the statements their AI agents make to customers.
The Root Cause
Air Canada deployed a chatbot without deciding who owned its outputs. The bot was treated internally as a deflection tool — a cost-saving overlay on top of existing policy documents. No one inside the company was tasked with reading what it actually told customers, comparing those answers against current policy, or correcting drift. The chatbot was a feature owned by the digital team, the policies were owned by revenue management, and the gap between them had no owner at all.
The second failure was the legal posture. By the time Moffatt’s case reached the tribunal, the airline’s instinct was to disown the chatbot rather than fix the underlying problem. That defense did not survive contact with a single small-claims judge. It also handed every plaintiff’s lawyer in the world a quotable precedent.
The Pattern to Watch For
Any AI system that speaks to customers in your company’s name is making representations on your behalf. If no human is reading those representations against your actual policies on a recurring basis, you are accumulating legal exposure with every conversation. The exposure is small per interaction and large in aggregate, which is exactly the shape of a liability that goes undetected until it appears in court filings.
What You Should Steal
Assign a single named owner for every customer-facing AI surface — chatbot, voice agent, email responder — and give that owner two weekly tasks: sample at least 50 conversations, and reconcile any policy statements the AI made against the policy of record. Cheap, boring, and the only thing that would have stopped Moffatt’s screenshot from becoming a global news story.