Air Canada and the Chatbot That Cost the Airline in Court

In February 2024, a small-claims decision in British Columbia became one of the most-cited AI stories of the year. It did not involve a data breach or a jailbreak. It involved a customer-service chatbot that gave a grieving passenger the wrong answer — and an airline that argued, unsuccessfully, that it should not be held responsible for what its own bot said.

After a death in the family, Jake Moffatt used Air Canada's website chatbot to ask about bereavement fares. The bot told him he could book a full-price ticket and apply for the bereavement discount retroactively within 90 days. That was wrong: Air Canada's actual policy did not allow retroactive bereavement claims. Relying on the bot, Moffatt booked, applied afterward, and was refused.

The argument that failed

When the dispute reached the British Columbia Civil Resolution Tribunal, Air Canada made a striking argument: that the chatbot was, in effect, a separate legal entity responsible for its own actions, and that the correct fare policy was available elsewhere on the site. The tribunal was unpersuaded. It held that Air Canada was responsible for all information on its website, whether it came from a static page or a chatbot, and that the company had made a negligent misrepresentation. Air Canada was ordered to pay damages and fees.

Why it matters beyond one airline

The sums were tiny; the principle is not. As companies rush customer-facing LLMs into production, the Air Canada ruling is the clearest early signal that the legal system will treat an AI assistant's statements as the company's own. A model that confidently invents a policy, a discount, or a warranty term is no longer just a product-quality problem — it is a liability the business owns.

There is a security dimension too. A chatbot that can be talked into stating a favourable policy — through leading questions or prompt injection — becomes an avenue for fraud and reputational damage. The same lack of grounding that produced an honest mistake here is what an attacker exploits deliberately.

The takeaway

• Ground customer-facing models in authoritative sources and constrain them from inventing policy. If the answer is not in an approved knowledge base, the bot should defer, not improvise.
• Treat every public generative surface as an official channel — reviewed, monitored, and owned by a team, not a bolt-on widget nobody governs.
• Know where these bots exist. Shadow customer-service assistants, stood up by a single team without oversight, carry the same liability with none of the controls.

Air Canada's chatbot did not get hacked. It simply spoke for the company, got it wrong, and the company paid. For every organisation deploying AI at the customer edge, that is the cheapest lesson on this list to learn from someone else.