Air Canada Learned the Hard Way: If Your Chatbot Promises It, You Owe It

In November 2022, a man named Jake Moffatt needed a last-minute flight from Vancouver to Toronto after his grandmother died. Before booking, he asked the chatbot on Air Canada’s website about bereavement fares. The bot told him to book now and apply for the discount within 90 days of travel. That policy did not exist. Air Canada’s actual rule requires bereavement fares to be approved before flying, and when Moffatt filed his claim, the airline refused it.

What happened next is why this case matters to every business running a customer-facing bot. Moffatt took Air Canada to British Columbia’s Civil Resolution Tribunal, and in February 2024 the tribunal ruled against the airline, ordering it to pay C$812.02 in damages and fees. The dollar amount is trivial. The reasoning is not.

The Defense That Failed

Air Canada argued it could not be held liable for information provided by its chatbot, suggesting the bot was effectively a separate entity responsible for its own statements. Tribunal member Christopher Rivers was unpersuaded. “This is a remarkable submission,” he wrote. “While a chatbot has an interactive component, it is still just a part of Air Canada’s website. It should be obvious to Air Canada that it 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 airline also argued that the correct policy was available on another page of its own site, and that Moffatt could have checked. Rivers rejected that too, noting that Air Canada never explained why customers should have to verify one part of its website against another. His conclusion: “I find Air Canada did not take reasonable care to ensure its chatbot was accurate.”

What The Ruling Actually Established

Strip away the aviation details and the tribunal established a principle that applies to a 40-person services company as fully as it does to a national airline: when your AI speaks to a customer, your company is speaking. There is no legal category for “the bot said it, not us.” A generated answer about pricing, policy, refunds, or terms carries the same weight as the same sentence typed by an employee, except the employee would have known the policy and the bot invented one.

The failure mode here was not a bad model. It was a missing gate. The chatbot was allowed to make policy statements to customers with no mechanism confirming those statements against the actual policy, and nobody had decided in advance what the bot was and was not allowed to commit to.

Before Your Bot Speaks For You

Inventory what your bot can promise. List every topic where a wrong answer creates an obligation: prices, refunds, policies, timelines, availability. That list defines where the risk lives.

Route commitments through a human gate. Anything containing a number, a policy, or a promise waits for human approval before a customer sees it, or the bot answers from retrieved policy text with a citation rather than generating from memory. Informational questions can flow freely. Commitments cannot.

Assume the transcript ends up in front of a judge. Moffatt screenshotted his conversation. Every customer can. Design the bot’s answers as if each one is a document your company may have to honor, because as of February 2024, it is.

The commitment gate, a human between the AI and any promise a customer can rely on, is the first control we design into every customer-facing deployment, and this ruling is the reason it is not optional.

Lessons from Bad AI Implementations is an ongoing series on what failed AI deployments teach operators. If you want to know which of these failure patterns your own workflows are exposed to, that’s what a Hiero workflow audit finds.