Mengchun’s Comments
This case is one of the early Chinese cases involving reputational harm caused by AI-generated search answers and further illustrates the developing judicial approach toward AI hallucination disputes in China.
The court distinguished between ordinary search suggestions and AI-generated answers. While search suggestions such as “related searches” were not found infringing, the court found that the relevant “AI Smart Answer” content was generated through Baidu’s AI processing and synthesis of text and images.
The case is significant because the court still found Baidu liable despite its arguments regarding AI hallucination and technological limitations.
At the same time, the judgment did not further elaborate on the specific standards for determining platform fault and the scope of platform duties in AI hallucination cases.
For more detailed discussion regarding China’s developing fault-based liability framework for AI hallucination cases, please also see our earlier analysis of the Hangzhou Internet Court’s landmark AI hallucination judgment:
I. Key Issue
When an AI-generated search answer falsely states that a specific lawyer was sentenced to prison and displays his professional photograph:
Should the AI service provider bear liability for defamation?
II. Court’s Interpretation
The court treated ordinary search-related suggestions and AI-generated answers differently.
The court held that Baidu Search provided information retrieval services and did not process or edit the relevant data or third-party webpage content. The court further noted that Baidu could not control third-party website content and that search engines were subject to updating cycles.
The court found that the search dropdowns, “people also search,” and related search terms were generated based on users’ search behavior and search records in order to facilitate future searches. Baidu itself did not process or edit those contents.
The court further held that search terms such as “Lawyer Li Xiaoliang,” “Nanjing Lawyer Li Xiaoliang,” “Jiangsu Lawyer Li Xiaoliang,” and “Has Jiangsu Lawyer Li Xiaoliang been imprisoned” did not themselves clearly point to the plaintiff.
On that basis, the court found that Baidu’s failure to proactively update or remove those search terms did not constitute subjective fault, but rather amounted to a lawful performance of its obligations.
The court also noted that Baidu could not directly block the relevant search terms, as doing so could negatively affect the rights and interests of other users.
Accordingly, the court found that Baidu’s dropdown suggestions, “people also search,” and related search functions did not constitute infringement against the plaintiff.
By contrast, the court found that Baidu’s “AI Smart Answer” content was generated through AI processing that combined text and image into a specific answer.
The AI-generated answer stated that “Lawyer Li Xiaoliang was sentenced to three years’ imprisonment” and referred to “defendant Li Xiaoliang” and “the crime of explosion.” The answer also displayed a photograph of Li wearing a lawyer’s robe.
The court held that this content clearly pointed to the plaintiff and contained obviously negative and damaging implications, which objectively lowered the social evaluation of the plaintiff.
III. Case Background
Li Xiaoliang is a practicing lawyer in Nanjing, Jiangsu Province.
In September 2024, he discovered that when users searched terms such as “Lawyer Li Xiaoliang,” “Nanjing Lawyer Li Xiaoliang,” and “Jiangsu Lawyer Li Xiaoliang” on Baidu’s mobile app and website, certain search suggestions and related search terms appeared, including whether he had been sentenced or involved in a criminal case.
On September 30, 2024, when searching “How many years was Lawyer Li Xiaoliang sentenced to,” Baidu’s “AI Smart Answer” generated the following response:
“Three years. Lawyer Li Xiaoliang was sentenced to three years’ imprisonment. According to the search results, defendant Li Xiaoliang was sentenced to three years’ imprisonment for the crime of explosion.”
The answer was accompanied by Li’s photograph in a lawyer’s robe.
Li sued Baidu, alleging that the false AI-generated content seriously damaged his reputation.
Baidu argued that the AI-generated answer was caused by technological limitations and the uncertainty of AI hallucination. It also argued that it could not predict how users would ask questions or what results the AI model would generate, and that it had taken remedial measures afterward.
IV. Court’s Holding
1. Ordinary Search Suggestions Did Not Constitute Infringement
The court held that Baidu Search provided information retrieval services and did not process or edit the relevant data or third-party webpage content. The court further noted that Baidu could not control third-party website content and that search engines were subject to updating cycles.
The court found that the search dropdowns, “people also search,” and related search terms were generated based on users’ search behavior and search records in order to facilitate future searches. Baidu itself did not process or edit those contents.
The court further held that search terms such as “Lawyer Li Xiaoliang,” “Nanjing Lawyer Li Xiaoliang,” “Jiangsu Lawyer Li Xiaoliang,” and “Has Jiangsu Lawyer Li Xiaoliang been imprisoned” did not themselves clearly point to the plaintiff.
On that basis, the court found that Baidu’s failure to proactively update or remove those search terms did not constitute subjective fault, but rather amounted to a lawful performance of its obligations.
The court also noted that Baidu could not directly block the relevant search terms, as doing so could negatively affect the rights and interests of other users.
Accordingly, the court found that Baidu’s dropdown suggestions, “people also search,” and related search functions did not constitute infringement against the plaintiff.
2. The AI-Generated Answer Was Different
By contrast, the court found that Baidu’s “AI Smart Answer” content was generated through AI processing that combined text and image into a specific answer.
The AI-generated answer stated that “Lawyer Li Xiaoliang was sentenced to three years’ imprisonment” and referred to “defendant Li Xiaoliang” and “the crime of explosion.” The answer also displayed a photograph of Li wearing a lawyer’s robe.
The court held that this content clearly pointed to the plaintiff and contained obviously negative and damaging implications, which objectively lowered the social evaluation of the plaintiff.
3. Baidu’s AI Hallucination Argument Was Rejected in This Case
Baidu argued that all AI systems may hallucinate and that the result was accidental and technically unpredictable.
The court acknowledged that the content had a certain degree of unpredictability due to the early stage of AI development and existing technical limitations.
However, the court still found that the relevant content was generated through Baidu’s AI processing, clearly pointed to the plaintiff, and objectively harmed his reputation.
Accordingly, Baidu’s conduct constituted defamation.
V. Judgment
The Nanjing Jiangbei New Area People’s Court ordered Baidu to issue a written apology to Li Xiaoliang.
The court did not support Li’s claims for economic loss or compensation for emotional distress because he failed to provide sufficient evidence of such losses.
Baidu appealed.
In March 2026, the Nanjing Intermediate People’s Court dismissed the appeal and upheld the first-instance judgment.
After the judgment became effective, Baidu did not voluntarily perform the judgment. Li Xiaoliang subsequently applied to the court for enforcement.
Court:
Shanghai Jinshan District People’s Court (First Instance)
Shanghai Intellectual Property Court (Second Instance)
Final Decision Date:
March 12, 2026
Case No.:
First Instance:
(2025) 沪0116民初2354号
(2025) Hu 0116 Min Chu No. 2354
Second Instance:
(2025) 沪73民终1189号
(2025) Hu 73 Min Zhong No. 1189
