Article by Erwyn Durman, partner; Manno Manavhela, associate; and Precious Maphupha, candidate attorney at Webber Wentzel
With the growing use of artificial intelligence and its integration into corporate systems, a key question arises: can a company be held responsible for misleading statements made by its AI chatbot to a customer?
This issue was addressed in Moffatt v Air Canada, a case concerning a refund for a bereavement fare. Although decided by a Canadian small-claims tribunal in 2024 and not binding in South Africa, Moffatt is instructive. As AI-driven customer service tools become commonplace, its reasoning on when a company must answer for its chatbot’s words offers useful lessons for the South African landscape.
Moffatt v Air Canada
Following the death of his grandmother in November 2022, Mr Moffatt used Air Canada’s website chatbot to research bereavement fares for an urgent flight. The chatbot advised that a reduced bereavement rate could be claimed retroactively, within 90 days of the ticket being issued, by submitting a refund application. Relying on this information, Mr Moffatt booked one-way flights between Vancouver and Toronto at standard rates and subsequently applied for partial refunds.
Air Canada rejected the refund request, pointing to a different webpage on its site that stated that bereavement fares could not be applied retroactively.
The Tribunal found Air Canada liable for negligent misrepresentation and, applying the principle that Mr Moffatt should be placed in the position in which he would have been had the misrepresentation not occurred, awarded CAD 650.88 in damages – less than the claimed amount of CAD 880.36.
Lessons from the judgment
The key issue was whether Air Canada could be held liable for inaccurate information generated by a chatbot on its website.
The Tribunal defined negligent misrepresentation as occurring “when a seller does not exercise reasonable care to ensure its representations are accurate and not misleading”.
Air Canada admitted that the chatbot had used “misleading words” but argued that it could not be liable for the chatbot’s statements, suggesting that the chatbot was effectively a separate legal entity responsible for its own statements.
The Tribunal rejected this, describing the submission as “remarkable”, and found that the chatbot was simply part of Air Canada’s website. It held that a consumer could not reasonably be expected to know which section of the airline’s website was accurate when contradictory information existed. As a service provider, Air Canada owed Mr Moffatt a duty of care and had breached the applicable standard by failing to take reasonable steps to ensure that the representations made on its website and by the chatbot were accurate.
South African perspective
South Africa’s law of delict already recognises negligent misrepresentation as a basis for liability and the Consumer Protection Act 68 of 2008 prohibits misleading representations to consumers. No reported South African case has yet tested these principles in the context of an AI chatbot.
Even so, the reasoning in Moffatt is one that could plausibly find favour in South African courts: a company cannot treat its chatbot as an autonomous actor detached from its business. If a chatbot misleads a client about the terms, procedures or exclusions of a product or service, liability will likely rest squarely with the company that deployed it.
This matters as chatbots become standard features of client-facing services in South Africa. The efficiency gains are real but so is the exposure. A single misrepresentation can trigger financial, reputational and regulatory consequences and that risk may extend to AI systems used by third-party service providers.
South African courts may well adopt reasoning similar to that of the Canadian Tribunal: consumers cannot reasonably be expected to distinguish between “official” and “automated” sources of information when both are presented on a company’s platform.
With digital adoption accelerating across South Africa, companies should implement governance frameworks, compliance checks and human oversight of AI outputs now rather than wait for the first local test case.
