Jake Moffatt asked Air Canada's website chatbot about bereavement fares after a death in his family. The bot told him he could book now and claim the reduced fare afterward, as long as he applied within 90 days of the ticket being issued. That was not Air Canada's policy, and when he applied, the airline said no.
In February 2024, British Columbia's Civil Resolution Tribunal ordered Air Canada to pay him $812.02. The airline's defense is the part every brand leader should read twice. It argued it could not be held liable for information provided by one of its agents, a chatbot included. Tribunal member Christopher Rivers summarized this as the suggestion that the chatbot was a separate legal entity responsible for its own actions, called it "a remarkable submission," and wrote: "It should be obvious to Air Canada that it is responsible for all the information on its website."
In April 2025 it happened again, this time to a company that sells AI. Users of Cursor, the AI code editor, were being logged out unexpectedly. A support agent named Sam emailed one of them that this was expected behavior under a new login policy. There was no new policy, and Sam was a bot. Some users cancelled their subscriptions. Co-founder Michael Truell posted on Reddit acknowledging an "incorrect response from a front-line AI support bot" and apologized for the confusion.
Brand teams have always treated support as an operations function, a cost center with a script. Once an AI agent runs it, support becomes the one channel where your brand speaks in full sentences, in the first person, to one customer about that customer's actual problem, with no copywriter anywhere near the reply.
Air Canada's bot sounded helpful and Sam sounded like a person, so tone was fine in both cases. Each one invented a rule. Most brand voice guides spend pages on tone (warm, plain, confident) and say nothing about the question that cost these companies money, which is what the voice is allowed to promise.
Full disclosure: I build one of these. Resolvly.ai is an AI support product for small businesses, so I have a commercial reason to tell you the problem is solvable, and you should weigh the next paragraph with that in mind. While building it, the hardest engineering question and the hardest brand question kept turning out to be the same one: what the agent says when it does not know. "I'm not sure, let me get you someone who is" is a brand line, and somebody has to write it, approve it, and make it sound like you.
The practical fix belongs in the brand guidelines, right next to the voice section. List which policies the agent may state, which it must quote word for word from a source page, which questions always go to a human, and the exact words it uses to admit it does not know. Label it as AI. Sam's human name is a large part of why a bug turned into a story about deception. Then review support transcripts the way you review ads, on a schedule, with someone who has the authority to change the rules.
Air Canada paid $812.02 to learn that its chatbot speaks for the brand, and most companies will pay more for the same lesson, with far less warning, the first time a customer screenshots a promise nobody at the company ever made.