REG-2024-02-14 · 14 FEBRUARY 2024 Verified

Air Canada — chatbot accountability did not stop at the interface

An automated assistant contradicted Air Canada’s published bereavement-fare policy. A customer relied on the answer, and the tribunal held the airline responsible for the representation made through its website.

Primary source
British Columbia Civil Resolution Tribunal · reviewed 6 August 2026
Moffatt v. Air Canada, 2024 BCCRT 149
canlii.org · 2024 BCCRT 149

CanLII reproduces the Civil Resolution Tribunal decision. The entry records the tribunal’s findings on negligent misrepresentation and organizational responsibility for information supplied through the chatbot.

What happened

The chatbot contradicted the policy it was representing.

In November 2022, following the death of a family member, Jake Moffatt used the chatbot on Air Canada’s website to ask about bereavement fares. The chatbot advised that a traveller who had already travelled could apply for a reduced bereavement rate within 90 days. Air Canada’s published policy stated that bereavement consideration did not apply after travel had been completed.

Moffatt relied on the chatbot’s answer, purchased the tickets and later requested the reduced fare. Air Canada denied the request and acknowledged that the chatbot had supplied misleading information. The Civil Resolution Tribunal found Air Canada liable for negligent misrepresentation and ordered compensation.

One channel, two answers

The conversational interface stated that a retroactive request was available. The linked policy page stated the opposite.

Reliance before correction

The customer acted on the chatbot’s representation before the contradiction was identified and corrected through human correspondence.

Organizational responsibility

The tribunal rejected the suggestion that the chatbot could be treated as separate from the organization responsible for the website and its information.

The amount at issue was modest. The governance principle was not: an organization does not transfer responsibility for customer-facing information by changing the interface through which the information is delivered.

This entry records a tribunal decision about one customer dispute. It does not generalize the legal result to other jurisdictions or systems.

Which control failed

The approved source and the conversational layer were not synchronized. The policy page and the chatbot were two representations of one organizational rule. No effective content control prevented them from contradicting each other.

Ownership was treated as a technology question instead of an organizational duty. The chatbot did not own the fare policy, approve exceptions or carry liability. An accountable business owner should have governed the source, the generated answer and the escalation path as one customer communication system.

Correction occurred after reliance. A control that depends on a customer discovering the contradiction and pursuing a dispute is not monitoring. Customer-facing assistants need source versioning, test cases for material policies, clear escalation and a process that withdraws unreliable answers before they are acted upon.

Zevana exposure analogue

Exposure floors are recorded, not averaged down.

The assistant was customer-facing and could create financial and contractual consequences, but the recorded incident involved a bounded fare dispute rather than a high-stakes eligibility, safety or rights determination. The analogue is moderate—and could rise quickly if the same control pattern affected many users or more consequential decisions.

Practice addressed
Atlas + Strata

Accountability, source & control readiness

Atlas establishes the policy owner, approval authority and responsibility for customer representations. Strata tests authoritative-source synchronization, version control, contradiction handling, monitoring and the human escalation route.

Who owns the answer when the interface is automated?
Ask your organization
One question, this week

Which approved source governs our chatbot’s answer, who owns that source, and what happens when the chatbot contradicts the organization’s published policy?

Test one material policy end to end. If the source, interface, human channel and exception process can give different answers, the organization has not defined one controlled position.

← All register entries
Begin with visibility

Ask it of your own systems before someone else does.

A short conversation about what your organization is running, who owns it, and what evidence exists. Scope and fee are confirmed after that conversation — never before.