Mengchun’s Comments
This case is notable because it shows how courts are beginning to deal with AI “hallucinations” in real commercial settings. Instead of looking only at what the AI said, the Shanghai Pudong Court paid close attention to how the conversation happened, what the user repeatedly asked, and what safeguards the platform had in place. The decision makes clear that when misleading AI outputs are driven by strategic user prompting, and the model provider has acted reasonably, courts may be reluctant to impose liability—offering practical guidance for AI platforms, brand owners, and their counsel.
I. Issues Presented
- Trademark Use
Does the automatic display of a third party’s trademark in an AI customer service interface—where such display is triggered by user-entered search keywords—constitute trademark use and trademark infringement? - Unfair Competition and AI Hallucinations
Where AI-generated responses contain inaccurate or misleading statements as a result of repeated or strategic user prompting, can such conduct give rise to liability under the Anti-Unfair Competition Law? - Liability of the Model Service Provider
Under a fault-based liability framework, may a large model service provider be held directly liable for user-induced AI outputs, and what constitutes fulfillment of its duty of reasonable care?
II. Case Background
The plaintiff lawfully held the right to use and enforce a registered hotel brand trademark with substantial market recognition.
Defendant One, an entity under the Huazhu Group, operated hotel franchise promotion services.
Defendant Two, the large model service provider, operated the “Bai某” App, which offered search engine functionality and AI agent services.
In February 2024, Defendant Two made available an AI-powered customer service system (“Qiaocang”) on its platform, which Defendant One used for hotel franchise consultations.
In May 2024, the plaintiff searched for “A Hotel franchise” (with “A Hotel” being the plaintiff’s brand) on the Bai某 App. The search results displayed a promotional link relating to Defendant One. After clicking the link, the user was directed to an AI-powered customer service interface made available on Defendant Two’s platform and used by Defendant One, where the search keywords were automatically carried over into the dialogue input field.
The plaintiff then engaged in multiple rounds of conversation with the AI system, repeatedly posing questions centered on “A Hotel franchise.” In response, the AI system generated statements such as “You can franchise A Hotel” and “We are a Huazhu brand, A Hotel.”
Throughout the interaction, the interface prominently displayed AI identifiers, Defendant One’s company name, and Defendant One’s own brand information.
The plaintiff alleged that the defendants, without authorization, used the plaintiff’s trademark to promote hotel franchising, thereby infringing trademark rights and engaging in unfair competition by free-riding on the plaintiff’s commercial goodwill.
III. Court’s Analysis and Holdings
A. Trademark Infringement
The court first examined whether the disputed references constituted “trademark use” within the meaning of the Trademark Law.
The court emphasized that the disputed references appeared exclusively within a clearly labeled AI customer service interface designated for Huazhu brand franchise consultations, which expressly defined the scope of the consultation, for example:
“Which Huazhu brand hotel franchise information would you like to consult?”
The plaintiff’s trademark was not independently emphasized or highlighted, and part of the disputed content merely reflected the user’s own search keywords being transferred into the consultation interface to facilitate dialogue.
In this context, the court held that the references served a descriptive and contextual function, rather than identifying the source of goods or services. Because the disputed use could not perform a source-identifying function, it did not constitute trademark use under trademark law.
Accordingly, the court rejected the plaintiff’s trademark infringement claim.
B. Unfair Competition and AI Hallucinations
The court next considered whether the defendants’ conduct violated Article 2 of the Anti-Unfair Competition Law.
The court observed that all disputed conduct occurred within a transparent and clearly defined consultation environment, and that the “query carry-over” feature merely imported the user’s own search terms into the consultation dialogue.
Importantly, the AI system’s responses were generated in a one-to-one conversational setting and were directly responsive to the plaintiff’s repeated inquiries regarding “A Hotel franchise.” Under these circumstances, a reasonable user would have understood the context, scope, and intent of the consultation.
Although the court acknowledged that the AI-generated responses were overly simplistic and potentially prone to misunderstanding, it emphasized that, considering:
- the specific conversational context,
- the user’s repeated and targeted prompting, and
- the current technical limitations of generative AI systems,
there was insufficient evidence of consumer confusion, deception, or disruption of market competition.
The court therefore held that the plaintiff failed to establish unfair competition.
C. Liability of the Large Model Service Provider
With respect to Defendant Two, the court applied Article 1165 of the PRC Civil Code, adopting a fault-based liability standard.
The court found no fault on the part of the model service provider, based on the following considerations:
- Purpose and Intent
The “query carry-over” function was designed to enhance user experience rather than to infringe trademark rights, and there was no evidence of actual harm to the plaintiff or consumers. - Reasonable Safeguards
Defendant Two had implemented reasonable precautions, including prominent AI identifiers, disclaimer clauses in service agreements, and mechanisms allowing intervention or adjustment. - User-Induced Output
Despite being aware that the consultation took place within a Huazhu-branded consultation interface provided by Defendant Two, the plaintiff’s representative repeatedly posed the same targeted questions, materially influencing the AI system’s responses and contributing to the alleged hallucinations. - Lack of Notice
The plaintiff had not provided prior notice or lodged any complaint with the defendants before initiating litigation.
Based on these factors, the court concluded that Defendant Two had fulfilled its duty of reasonable care and did not constitute direct infringement.
IV. Practical Significance
This decision establishes an important judicial principle: AI liability cannot be assessed in the abstract, divorced from user behavior and contextual transparency. Where AI outputs are driven by strategic or repetitive prompting, and the service provider has implemented reasonable safeguards, courts may decline to impose liability.
For brand owners, the case underscores the evidentiary burden required to prove trademark use or unfair competition in AI-mediated environments. For AI platforms and developers, it highlights the practical value of clear interface disclosures, technical guardrails, and effective complaint-handling mechanisms in mitigating legal risk.
As courts increasingly confront disputes involving generative AI, this judgment offers a nuanced, fault-based framework likely to influence future AI governance and litigation strategies in China.
Court:
Shanghai Pudong New Area People’s Court
Final Decision Date:
October 31, 2025
Case No.:
(2024)沪0115民初95826号
(2024) Hu 0115 Min Chu No. 95826
