Chatbot Liability Ruling Sticks Air Canada With Bill

An 812.02 Canadian dollar award against Air Canada turned a bad support answer into the clearest chatbot liability warning yet: if your bot speaks through your customer channel, courts may treat it as your company speaking.

That is the practical lesson from Moffatt v. Air Canada, 2024 BCCRT 149, where the British Columbia Civil Resolution Tribunal held the airline liable after its chatbot invented a bereavement fare rule that did not exist, according to PYMNTS. The dollar amount was small. The governance signal was not.

For executives, the issue is no longer whether a customer-facing AI chatbot can make mistakes. It can. The harder question is whether the business can distance itself from those mistakes after customers rely on them. In Air Canada’s case, the tribunal’s answer was blunt: no.

Why did an $812.02 chatbot error become a chatbot liability warning?

Jake Moffatt used Air Canada’s chatbot after his grandmother died. The bot told him he could buy a full-fare ticket and claim a retroactive bereavement discount within 90 days. That policy did not exist.

When Air Canada denied the claim, it argued that the chatbot bore responsibility for its own statements, the American Bar Association reported at the time, per PYMNTS. Tribunal member Christopher Rivers rejected that framing.

“In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions,” Rivers wrote. “This is a remarkable submission.”

The tribunal ordered Air Canada to pay 812.02 Canadian dollars (about $570) in damages and fees. More important, it treated the chatbot as part of Air Canada’s website, not as some independent actor floating outside the company’s control.

That is the core chatbot liability shift. A chatbot answer can become a business statement when it appears inside an official service channel and a customer reasonably relies on it.

This is also why the case matters beyond airlines. Banks, retailers, platforms, insurers and software companies are all pushing AI into support flows. The source material does not establish how courts will treat every sector, but Air Canada shows the starting risk: companies may own what their automated agents tell customers.

For readers tracking adjacent AI control questions, XOOMAR’s Runaway AI Spending Forces a Return to Cloud Controls and Anti-Vaccine Myths Cluster Around AI Chatbot Users are useful companion reads.

How hallucinations differ from confabulations when customers rely on them

AI hallucination is the broader false-output problem: a fake citation, a wrong number, an invented person, or a made-up instruction presented as fact.

AI confabulation is more specific. It happens when the system does not know the answer and fills the gap with something that sounds plausible. The danger is tone. The answer can read like a real company policy because the bot states it with confidence.

That is what makes confabulated customer-service language so risky. A fabricated refund rule, subscription limit, fare condition or account instruction can look official when it appears in a branded chat window.

The Air Canada case involved a bot inventing a retroactive bereavement fare process. The Cursor incident in April 2025 followed the same pattern in a different market. An AI support bot named Sam told a developer that Cursor had a new policy limiting each subscription to one device. The policy did not exist.

Cursor Co-Founder Michael Truell corrected the record on Reddit:

“Hey! We have no such policy,” Truell said. “You’re of course free to use Cursor on multiple machines. Unfortunately, this is an incorrect response from a front-line AI support bot.”

Before that correction reached users, the invented policy had circulated on Hacker News and Reddit, according to PYMNTS. That is the operational problem. A bot does not need legal authority to create customer confusion. It only needs distribution.

When does a bot answer start to look like a company statement?

The Air Canada ruling turned on negligent misrepresentation, according to related reporting cited in the supplied material. The tribunal found that Air Canada “did not take reasonable care to ensure its chatbot was accurate,” and rejected the idea that customers should have to double-check one part of the airline’s website against another.

That finding is narrower than saying every chatbot mistake automatically creates liability. It does not. The stronger pattern is reliance.

A customer has a clearer complaint when they act on a bot’s answer and lose money, miss a deadline, accept worse terms, or give up a right. In Moffatt’s case, the bot’s statement shaped how he understood the fare process after a death in the family.

XOOMAR analysis: the same factual pattern would be most sensitive in channels where the answer changes a customer’s financial position. Think checkout flows, banking portals, insurance claims, travel bookings, loan servicing, or account support. The supplied sources do not prove how each industry’s cases would come out, but the Air Canada logic is easy to follow: the more official the bot looks, the harder it becomes to call the output mere software noise.

Disclaimers are not a magic escape hatch in the facts we have. The key Air Canada point was that the chatbot was embedded in the airline’s own website, and the tribunal saw no reason customers should treat another webpage as more trustworthy than the chatbot.

Air Canada shows the evidence pattern courts can understand

The Air Canada dispute was easy to grasp because the record had a concrete chain:

That is the model companies should study. Courts do not need to understand every detail of model architecture to evaluate a misleading customer interaction. They can look at the transcript, interface, branding, links, policy text and the company’s response after the dispute.

The Cursor episode adds another lesson. Even without a cited court ruling, a false support answer can spread fast when users believe it describes real product terms. In Cursor’s case, the correction came from a co-founder after the invented device-limit policy had already circulated.

That reputational path matters because chatbot liability is not only litigation risk. PYMNTS also points to insurers and regulators beginning to treat AI errors as financial exposure.

Insurers and regulators are starting to price the risk

In May 2025, Lloyd’s of London launched an insurance product covering AI hallucination-related losses, offered through Armilla. PYMNTS says those policies cover court claims against a business if a customer or third party is harmed by an underperforming AI product.

FINRA also flagged hallucinations in its 2026 Annual Regulatory Oversight Report as a compliance concern for broker-dealers. The regulator warned firms to develop procedures for AI agents that may act beyond the user’s intended scope, and defined hallucinations as instances where a model generates inaccurate or misleading information while presenting it as factual.

Then Scaled Cognition raised $100 million in June to build enterprise-grade hallucination controls. The funding detail matters because it shows a market response to the same problem courts are seeing: bad AI answers are no longer just product defects. They can become claims, compliance issues and insurance events.

How companies can redesign bots before bad answers become exhibits

Companies do not need perfect chatbots. They need bots that know when to stop.

The controls flow directly from the failures in the source material:

• Scope limits: Keep bots away from high-risk promises involving refunds, fees, deadlines, eligibility, regulated disclosures and anything that changes a customer’s financial position.
• Approved sources: Tie answers to current company policies and prevent the model from inventing policy language when no source supports it.
• Escalation rules: Route sensitive questions to humans instead of letting the bot improvise.
• Uncertainty handling: Make “I don’t know” a valid outcome, not a failure state.
• Audit trails: Preserve transcripts, interface versions, policy sources and incident reviews so the company can show how the system behaved.
• Testing: Red-team for fake policy claims, refund errors, deadline mistakes and confident answers to questions outside the bot’s knowledge base.

The forward-looking issue is not whether courts will understand AI well enough to assign responsibility. Air Canada suggests they may not need to. If the chatbot sits in the company’s channel, gives a polished answer, and the customer relies on it, the next dispute may start from a simple premise: the business owns the words its bot delivered.

Impact Analysis

• Courts may treat chatbot responses as official company statements when customers rely on them.
• The Air Canada ruling shows even small AI errors can create legal and reputational risk.
• Businesses using customer-facing AI need stronger controls, oversight, and accountability.