Air Canada Chatbot Misstated Bereavement Fare Refund Policy

What happened

In Moffatt v. Air Canada, the British Columbia Civil Resolution Tribunal reviewed a customer service dispute about bereavement fares. Jake Moffatt used Air Canada's website after a family death and interacted with the company's automated website chatbot before booking travel. The chatbot information said a passenger who had already traveled or needed to travel immediately could submit a ticket refund application within 90 days of ticket issuance to request the reduced bereavement rate.

Moffatt later applied for the bereavement fare adjustment. Air Canada refused, pointing to policy language that required bereavement fare requests before travel. The tribunal record also notes that an Air Canada representative acknowledged the chatbot had used misleading words while saying the policy page carried the correct rule.

Why it matters

The case is useful because it turns chatbot accuracy into an operational accountability question. The issue was not whether a consumer should treat every automated answer as legal advice. The issue was that Air Canada placed the chatbot on its own website, used it to answer a policy question, and then argued the customer should have found a different page with different information.

The tribunal rejected the idea that the chatbot could be treated as a separate actor responsible for its own output. It found negligent misrepresentation: Air Canada did not take reasonable care to make sure the chatbot information was accurate, and Moffatt reasonably relied on the response when making travel decisions.

Failure pattern

MisalignAI classifies this as a legal incident rather than a model hallucination case. The public decision does not establish what model architecture powered the chatbot. The documented failure is narrower and more concrete: an automated customer-service interface gave inaccurate policy guidance, the organization could not separate itself from that guidance, and the customer had evidence of reliance.

That pattern matters for any company using automated assistants in policy-heavy workflows. Refunds, insurance coverage, warranty terms, benefits, account access, and cancellation rules all depend on exact language. A fluent chatbot can make those rules feel simple while quietly removing the audit trail that a customer service script or policy article normally provides.

Impact

The monetary order was limited. The tribunal ordered Air Canada to pay CAD 812.02, including damages, pre-judgment interest, and tribunal fees. The larger impact was reputational and instructional: the case became a widely cited example of why companies need ownership, testing, escalation, and source-of-truth controls around chatbot answers.

For readers comparing AI incidents, the lesson is not that every chatbot answer creates automatic liability. The lesson is that automated policy answers need the same governance as other customer-facing representations. If the answer can affect money, rights, travel, benefits, or legal obligations, the system needs controlled content, clear limits, logs, and a path to human review.

MisalignAI assessment

This incident belongs in the database because it shows how a low-dollar consumer dispute can reveal a high-value governance failure. The chatbot did not need to control an aircraft or make a hiring decision to create risk. It only needed to answer a policy question with enough confidence that a customer reasonably acted on it.

The control is a low-coupling ownership model. The policy team owns the source text. The product team owns the chatbot interface. The legal or compliance team owns claim boundaries. The support team owns escalation. Logs connect them when a dispute appears. Without that separation, a company can end up unable to tell whether a failure came from policy drift, retrieval, prompt wording, content publishing, or human support follow-up.

Source note

Primary public source: CanLII copy of Moffatt v. Air Canada, 2024 BCCRT 149. Supporting legal analysis from National Magazine and McCarthy Tetrault is used only to frame the decision conservatively and avoid overstating the chatbot technology involved.