Facts
The claimant registered with a generative AI app in March 2025, accepting its user agreement. On 29 June 2025 he asked about a university's admissions and the app gave inaccurate information about one of its campuses. When he tried to correct it, the AI insisted the campus existed and offered to pay the user 100,000 yuan if the content was wrong. Shown the official information, the app admitted the error and suggested he sue at the Hangzhou Internet Court. He claimed 9,999 yuan from the company that developed and ran the app. The company argued that AI output is not a declaration of intent, that it had met its duty of care and that he had suffered no loss.
Question
The court had three questions to answer. First, whether a compensation 'promise' produced by the AI on its own during a chat could count as the provider's declaration of intent. Second, whether inaccurate output from a generative AI service attracts strict product liability or ordinary fault-based liability. Third, if fault applies, what the duty of care of such a provider consists of and whether the defendant had met it. The court also examined whether the claimant had suffered the loss he alleged and whether the false information was an adequate cause of it.
Decision
The court held that AI has no civil-law personality and cannot declare intent, and that the defendant had shown no intention to be bound by AI output. It classed generative AI as a service rather than a product and applied general fault liability under Article 1165(1) of the Civil Code. It described the duty of care in three layers: strict review for toxic, harmful and unlawful content; clear, prominent notice of the service's limitations; and industry-standard technical measures such as retrieval-augmented generation to improve accuracy. As the defendant had given notices on its welcome page, user agreement and interface and used such measures, it was not at fault. Loss and causation were unproven; the claim failed and neither party appealed.
Why it matters
Press reports described it as China's first tort case over an AI 'hallucination'. The judgment says plainly that commitments a chatbot generates on its own do not bind the operator, a point that invites comparison with Moffatt v. Air Canada, where a company was held responsible for its website chatbot's wrong information. The three-layer duty of care indicates that inaccurate output alone does not create liability, while missing warnings or technical safeguards may amount to fault. The court also indicated that prompt warnings may be needed in high-stakes uses.
Related stages
The court publicised the case on 29 December 2025; the judgment date and case number could not be verified. The Nanjing case in which Baidu was held liable for a defamatory AI answer is listed separately as an example where the fault analysis led to a different outcome.