Air Canada and the Chatbot Nobody Owned
Air Canada's website chatbot told a grieving customer he could claim a bereavement discount after he flew. The policy said otherwise. A tribunal held the airline to what its bot had promised.
A customer-facing chatbot gave answers nobody checked against the real policy, and no one owned what it said. The control absent: treating the bot's output as the company's own word, with review, limits on consequential topics, and a clear line of accountability.
In November 2022, Jake Moffatt booked a last-minute Air Canada flight after a death in his family. Before booking, he asked the airline’s website chatbot about bereavement fares. The bot told him he could buy a regular ticket and apply for the bereavement discount within ninety days, after travel. He did exactly that.
The bot was wrong. Air Canada’s actual policy, on a page the bot even linked to, said bereavement fares could not be claimed after the flight. The airline refused the refund.
Moffatt took it to the British Columbia Civil Resolution Tribunal. In February 2024, the tribunal ruled against Air Canada. The airline had argued, remarkably, that the chatbot was a separate legal entity responsible for its own actions. The tribunal rejected that outright: the chatbot was part of Air Canada’s website, and the airline was responsible for everything on it, the bot included. It ordered Air Canada to pay.
Where it failed
The amount was small, a few hundred dollars. The principle is not. A probabilistic system generated a plausible, confident, wrong answer, which is what these systems do some of the time. The real failure came after that: nobody owned the error.
The chatbot’s answers were never checked against the policy they claimed to represent. There was no review, no constraint tying the bot to the ground truth, and, when it mattered, no acceptance that the company was accountable for what it had said. The break was in process and technology: a generative system put in front of customers to speak for the airline, with no one standing behind its words until a tribunal made them.
A company owns what its chatbot tells a customer. It took a tribunal to make Air Canada accept that.
How it could have been caught
Treat the output of a customer-facing system as the company’s own statements, because legally and practically that is what they are. Tie the bot to the real policy, so it quotes the source rather than improvising near it, and constrain it from inventing terms on consequential topics like refunds and fares. Give it a clear owner, someone accountable for what it tells customers, and a way for a person to catch and correct a wrong answer before it becomes a promise. The cheapest version is simple: do not let a generative system free-form answers on questions where being wrong creates a liability.
What it means for AI
Everyone deploying a customer-facing AI, which is fast becoming everyone, will eventually meet this ruling. The appeal of these bots is that they answer anything in natural language. That is also the risk: they will answer anything, including confidently and wrongly, and the organization is on the hook for what they say. You cannot deploy the helpfulness and disown the mistakes.
Ownership is the entire question. Decide who is accountable for what the system says before you put it in front of customers, and bind it to the policy it is supposed to represent. The rule the tribunal had to state out loud is the one to plan around: a company owns what its AI tells the public, and every court and regulator that follows will start from the same place.
← Back to the Atlas