Air Canada’s $812 AI Debacle: The Case Every Business Owner Deploying AI Must Reckon With
How one hallucinating chatbot, one bereaved customer, and one watershed tribunal ruling permanently rewrote the jurisprudence of AI accountability in commerce.
What happened — Moffatt v. Air Canada
One consequential misstep, one bereaved passenger, and a ruling the world watched
In November 2022, Jake Moffatt found himself in a distressing bind. His grandmother had just passed away in Ontario and he needed to book a last-minute flight. He visited Air Canada’s website and — as millions of customers do daily — consulted the airline’s AI chatbot for guidance on bereavement fares.
What the chatbot told him was fictitious. It confidently instructed him to purchase a full-price ticket and submit a bereavement discount application within 90 days of the flight. He trusted it. He booked. He attended his grandmother’s funeral. He submitted the application — exactly as instructed. Air Canada refused, pointing to its actual policy on a different webpage: bereavement consideration had to be requested before travel, not after.
Case chronology · Moffatt v. Air Canada · 2024 BCCRT 149
November 2022
The chatbot conversation
Moffatt queries Air Canada’s AI chatbot about bereavement fares. The chatbot fabricates a post-travel application policy that does not exist in Air Canada’s official documentation.
November – December 2022
Application submitted and denied
Moffatt returns from the funeral and applies per the chatbot’s instructions. Air Canada refuses, acknowledges the chatbot used “misleading words,” but declines any refund — citing a separate webpage Moffatt had not seen.
December 2022 – February 2024
Dispute escalates to tribunal
After months of fruitless correspondence, Moffatt files a claim of negligent misrepresentation with the British Columbia Civil Resolution Tribunal — Canada’s small-claims adjudicator.
February 14, 2024
Watershed ruling — Moffatt wins
Tribunal Member Rivers rules for Moffatt. Air Canada ordered to pay CAD$812.02. The ruling generates international coverage in the BBC, Washington Post, and dozens of global legal publications.
The defence — and its spectacular failure
Air Canada’s audacious argument: “the chatbot is a separate legal entity”
Air Canada’s legal strategy was nothing short of audacious. The airline advanced three arguments: it bore no responsibility for its chatbot’s statements; customers should have cross-referenced a different part of the same website; and — most extraordinarily — the chatbot itself was a separate legal entity responsible for its own actions.
Every argument disintegrated before the tribunal.
Official ruling · 2024 BCCRT 149 · Tribunal Member Rivers
“In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission. While a chatbot has an interactive component, it is still just a part of Air Canada’s website. Air Canada is responsible for all information on its website.”
— TRIBUNAL MEMBER RIVERS, FEBRUARY 14, 2024
On the “check the other webpage” defence, the tribunal was equally unsparing. The ruling held that it was wholly unreasonable to require customers to corroborate information found in one section of a company’s own website against another. A chatbot bearing a company’s branding carries apparent authority — just as a printed brochure or a customer service agent does.
The jurisprudential principle established: chatbot outputs are company speech. Under apparent authority doctrine, companies cannot disclaim responsibility for their AI tools’ statements simply because the AI made a mistake. “The AI got it wrong” is not a defence in any tribunal, court, or consumer protection framework operating today.
“Air Canada does not explain why customers should have to double-check information found in one part of its website on another part of its website.“
— CIVIL RESOLUTION TRIBUNAL · MOFFATT V. AIR CANADA · 2024 BCCRT 149
Root cause — the anatomy of failure
What genuinely went wrong: AI hallucination meets governance vacuity
The $812 figure is almost incidental. What matters is the pathology of the failure — because every element of it is latent in thousands of businesses deploying AI customer service tools right now.
Air Canada deployed a generative AI chatbot — a large language model that constructs responses from training data, producing text that sounds authoritative and confident regardless of whether the underlying information is factual. Unlike rule-based systems that retrieve verified answers, generative models confabulate. And in this case, no human caught the confabulation before it reached a grieving customer at his most vulnerable.
| 01 No accuracy verification system. | The chatbot generated a non-existent policy and no automated or human check intercepted it before it reached a customer |
| 02 No escalation pathway. | The query was sensitive, time-critical, and emotionally charged — precisely thegenusof interaction requiring human review. None was available. |
| 03 No policy grounding or guardrails. | A well-governed chatbot retrieves from a verified, human-approved knowledge base. This one generated freely from model weights. |
| 04 No output monitoring. | Thespuriousbereavement policy circulated undetected until a tribunal ruling forced the issue into public view. |
| 05 No incident response protocol. | When Moffatt provided evidence, Air Canada’s response was peremptory denial — not acknowledgment, remedy, or system correction. |
In April 2024 — weeks after the ruling — Air Canada’s chatbot quietly absconded from the website entirely. A telltale admission that the deployment had never been adequately superintended from the outset.
Global business impact
Why this AI legal precedent is seismically important beyond $812
The financial award was trifling. The legal, reputational, and regulatory aftermath was not. This ruling has been cited by the American Bar Association, leading law schools, and legal practitioners from London to Singapore. Its reverberations reshaped how regulators, insurers, and general counsel think about every AI deployment in every customer-facing context.
$812
Award — fare diff plus tribunal fees
6+
US STATES · 2025
New AI chatbot liability laws enacted post-ruling
79%
INDUSTRY · 2025
Orgs using AI in customer experience — most without governance
84%
DELOITTE · 2026
Of AI-deploying businesses have NOT redesigned jobs around AI outputs
Colorado’s AI Act — effective 2026 — is now the most comprehensive state-level AI legislation in the United States, with statutory obligations around consumer disclosure, accuracy, and duty of care that map directly to the principles articulated in this ruling.
Lessons for every business deploying AI
6 AI governance imperatives every business owner must enshrine today
The Air Canada case is not an anomaly. It is a harbinger. Every business deploying a customer-facing AI tool in 2026 is operating within the legal landscape this ruling created.
01 Ground AI in verified policy
Restrict chatbot responses to retrieval from a human-approved knowledge base. Never allow generative models to construct policy answers from training weights.
02 Build human escalation paths
Sensitive, high-value, or emotionally charged queries must route to a human reviewer before any AI response is delivered to the customer.
03 Monitor outputs continuously
Audit chat logs. Track accuracy drift. Build detection systems that flag aberrant responses before a customer is harmed.
04 Train your team on AI liability
Before deployment, train employees to verify AI outputs, identify confabulation, and escalate anything that seems inconsistent with official policy.
05 Disclose AI interactions clearly
Customers must know when they are interacting with AI — not a human. This is now statutorily mandated in multiple US states and best practice everywhere else.
06 Document an incident response plan
When your AI errs — and it will — have a codified process for acknowledgment, customer remedy, and system correction. Peremptory denial is not a strategy.
The solution — human-led AI oversight
Human-in-the-loop AI is not a cost centre — it is your indemnity shield
The Air Canada ruling did not establish that AI is too mercurial to deploy. It established that AI deployed without qualified human oversight is a contingent liability. The businesses getting this right in 2026 are not choosing between AI and people — they are building symbiotic workflows where trained professionals govern, verify, and translate AI output into defensible, accurate, accountable business outcomes.
One chatbot. One bereaved customer. A CAD$812 ruling now canonical in AI legal discourse from British Columbia to the British Bar Association. Every business deploying AI in 2026 is operating inside the legal reality this case created. The question is whether you have built the human infrastructure to operate safely within it — or whether you are waiting to find out the hard way.
Deploy AI with confidence. We build the human oversight layer that keeps you protected, compliant, and accountable.
From AI output monitoring and knowledge base governance to escalation systems and compliance documentation — our pre-vetted virtual assistants, IT-BPM professionals, and developers provide the human-in-the-loop infrastructure that makes your AI deployment legally defensible and commercially reliable. Serving clients across the US, UK, and Australia.
-
AI output monitoring & quality control -
Knowledge base governance & SOPs -
Human-in-the-loop escalation systems -
AI compliance documentation -
Managed AI customer service teams -
Custom AI workflow development
50–80%
cost savings vs. US, UK, or Australian hires
with full AI governance capability built in