Article: Air Canada was ordered to pay $812.02 CAD after a tribunal found the airline liable for a chatbot that promised a refund that never existed.

How the dispute unfolded

In 2022 Jake Moffatt needed a last-minute flight for a funeral. He asked the airline’s online chatbot about bereavement fares. The bot replied that he could buy a full-price ticket and later request a refund. Relying on that answer, Moffatt paid $1,630.36 CAD. When he asked for the promised refund, Air Canada denied the request.

Moffatt took the matter to a tribunal. The adjudicators said the airline had engaged in negligent misrepresentation – the chatbot’s statement was a false promise that induced a purchase. They rejected Air Canada’s claim that a software agent cannot be a contracting party, noting that the bot lives on the company’s public website and therefore acts on the airline’s behalf. The tribunal ordered the airline to reimburse Moffatt $812.02 CAD.

Why the case matters

Most public outcry over AI errors focuses on obvious failures—data loss, broken code, or overtly biased outputs. This decision spotlights a subtler danger: an AI that sounds confident while inventing policy details can create enforceable obligations. The bot did not “delete” anything; it simply gave a clear, unqualified answer that a reasonable customer would treat as the airline’s official stance.

The ruling sets a legal precedent that AI-generated statements are not insulated by the “software is just a tool” defense. If a virtual assistant provides information a consumer relies on to make a purchase, that information may be treated as a contractual term. Companies that deploy chatbots without strict controls risk similar liability.

The technical root of the mistake

The error stemmed from a retrieval-augmented generation (RAG) pipeline. The system found the correct policy document but pulled the wrong sentence and presented it as fact. Guardrails that filter out gibberish or offensive language often miss well-written but inaccurate excerpts. In this case, the bot’s fluency—its lack of hesitation or “I’m not sure”—gave the false promise an air of authority.

What airlines and other firms can do now

  • Hard-code any financial or entitlement wording. Use fixed templates for statements about refunds, discounts, or fees.
  • Treat the chatbot as a frontline employee. Assume it can “sign” contracts in the eyes of the law and subject its outputs to the same review process as a human agent.
  • Separate search from generation. When the task is to locate policy text, return a link or a quoted excerpt rather than a paraphrased answer.

These steps may make the interaction feel less conversational, but they stop the kind of confident misinformation that cost Air Canada a court-ordered payout.

Counter-argument: Is the liability fair?

Air Canada argued that a piece of code lacks legal personhood and cannot be bound by contract. The tribunal’s rejection rests on the principle that the company presents the chatbot as an official point of contact. Critics worry that extending liability to AI could stifle innovation, forcing firms to over-engineer safety nets that diminish the user experience. The balance between protecting consumers and preserving AI agility will likely be debated in future cases.

What to watch next

  • Follow-up litigation. If Air Canada appeals, higher courts may refine the scope of AI-contract liability.
  • Regulatory guidance. Consumer protection agencies may issue rules on AI-generated disclosures, especially in finance, travel, and health sectors.
  • Industry standards. Tech groups could develop best-practice frameworks for “contract-safe” AI responses, similar to existing security and privacy guidelines.

The takeaway is clear: an AI chatbot that can answer questions about price, refund or entitlement is, in the eyes of the law, a representative of the business. Companies that treat it like a simple search box run the risk of paying out for promises their policies never intended to make.