China’s Supreme People’s Court issued guidance on AI-related disputes

On 7 September 2026, the Supreme People’s Court of the People’s Republic of China (SPC) issued the Opinions on Adjudicating AI-Related Dispute Cases in Accordance with Law, addressing issues such as AI infringement of personality rights and assessing copy right infringement in AI-generated content. 

The Opinions contain 24 provisions organised into five parts and are intended to guide courts in applying existing PRC laws to disputes involving rapidly developing AI technologies. 

The Opinions do not constitute standalone AI legislation. Instead, they provide adjudicative guidance for courts across China under existing legislation, including the PRC Civil Code, Cybersecurity Law, Data Security Law, Copyright Law, Anti-Unfair Competition Law, Consumer Rights Protection Law, Personal Information Protection Law and Civil Procedure Law. Their stated approach combines support for innovation with protection of rights and a baseline of safety while allocating responsibility by reference to factors such as the relevant application scenario, technical transparency, risk, control and preventive measures. 

In practice, the Opinions could serve important legal and policy guidance to AI developers, service providers, users, manufacturers, sellers, open-source developers and other parties where their AI-related activities become the subject of proceedings before a PRC court. 

Key takeaways

Businesses should be alert to the following in the Opinions: 

• Fault-based liability and notice-and-action liability on AI service providers: The Opinions expressly confirm that AI-related tort claims are governed by the fault principle unless legislation specifically provides for strict liability or a presumption of fault. This is important in the context of generative AI since the Supreme People's Court indicates that generative AI services, which generate or provide information through interaction with users, do not exhibit the type of inherent danger to life, health or property traditionally associated with ultrahazardous activities and should not ordinarily be subject to a no-fault liability regime. Instead, liability should be assessed by reference to fault, accounting for factors such as the AI system’s application scenario, degree of autonomy, technical and information transparency, potential risk and reach, measures taken by developers and providers, and a user’s ability to foresee and control harm. For generative AI content that infringes personality rights, an AI service provider that receives a notice and then fails to take necessary measures that meet the specified requirements may be liable for resulting harm. A user who maliciously induces infringing output may also be liable.
   
• Copyright infringement framework for AI-generated content: A notable feature of the Opinions is its establishment of a structured framework for assessing copyright infringement claims involving AI-generated content. Rather than adopting a presumption of liability against either rights holders or AI developers, the Opinions adopt a relatively balanced allocation of evidentiary burdens between the parties. courts are directed to assess factors such as the service type, industry characteristics, source of training data, each party’s participation, necessary measures and profit. A developer raising a non-infringement defence may be required to substantiate it with evidence concerning training-data sources, training records, model operation and relevant scientific theory. This approach seeks to address the information asymmetry inherent in AI systems without imposing an undue evidentiary burden on developers. The Opinion also confirms that users may be held liable where they knew or should have known of a pre-existing work and generated substantially similar content without a reasonable justification.

At the same time, the Opinions expressly refrain from taking a position on whether AI-generated content is copyrightable or whether unauthorised use of works for large-model training represents infringement. The Opinions also introduce more tailored evidentiary review for AI-related disputes, including scrutiny of the authenticity and integrity of electronic evidence and the technical factors underlying AI-generated content.

The Opinions addressed the following scenarios of AI-related disputes: 

• Protection of personality and privacy rights extending to digital likeness and synthetic voices: The Opinions expressly recognise that a natural person's personality rights may extend to an identifiable digital avatar or virtual persona capable of representing and identifying that individual. Against that backdrop, the Opinions address AI-enabled infringements involving digital likenesses, synthetic voices and “doxxing”. Unauthorised generation and use of an identifiable person’s digital likeness or synthetic voice may infringe personality rights while AI-assisted tracking or analysis of publicly available information to obtain or disclose private information may constitute an infringement of privacy.
   
• Using public personal information as training data may be permissible: Processing self-disclosed or otherwise lawfully public personal information within a reasonable scope for model training will generally not be treated as infringing personal information rights where the individual has not expressly objected. Consent must nevertheless be obtained where the processing has a significant impact on the individual’s rights and interests. Courts should assess purpose, necessity, appropriateness, data type and sensitivity, potential impact, the context of disclosure and the expected scope of use. Public availability alone does not make personal information unrestricted training material.
   
• Product liability applies to physical AI Products rather than standalone AI services: Product liability applies to AI products carried by a physical object. In assessing whether an AI product is defective and causing unreasonable harm and danger to persons and property, courts should consider purpose, self-learning capability, updates, user control, applicable standards, and whether limitations and foreseeable risks were truthfully explained and clearly warned by manufacturers and distributors.
   
• Consumer protection and algorithmic discrimination: The Opinions expressly regulate AI-enabled practices that harm consumer interests, including "big data price discrimination" where algorithms are used to impose unreasonable differential pricing or other transaction terms on consumers. Merchants engaging in such practice may be liable in tort where such practice unfairly restricts or impairs consumers' rights and interests. The Opinions further provide that fraudulent AI-enabled celebrity impersonation in consumer transactions may support claims for punitive damages. 
   
• Automated and assisted-driving vehicle accidents: The Opinions expressly address compensation liability for traffic accidents involving automated vehicles and vehicles equipped with assisted-driving functions. Courts may require data controllers to provide authentic and complete event records within the necessary scope, and defects may expose producers or sellers to liability alongside driver fault. 
   
• Liability framework for open-source AI components: For cases involving open-source software, courts should assess the liabilities of the open-source software developer, provider and downstream developers and providers by reference to the licence, restrictions, security and compliance measures, and disclosure. A developer or provider that makes certain code modules, required for the development of AI software, available on a free and open-source basis, and publicly explains their functions and security risks may not bear tort liability where another party’s use of such code modules has been deemed to cause infringement.
   
• Patentability of AI-related inventions: AI-related inventions that resolve technical problems should be protected as patents. A natural person who makes a creative contribution to the substantive features may be recognised as the inventor. Contract liability should reflect the agreement, the characteristics of AI research and development, and whether the developer used reasonable efforts. Lawfully obtained data interests, qualifying compilations, copyright works and trade secrets are also recognised as protectable interests under the applicable laws. 
   
• AI-generated litigation materials must be verified and disclosed: Parties and other litigation participants must do the following – verify the authenticity and accuracy of AI-generated pleadings, case-search reports and similar materials before filing, disclose the use of AI assistance to the court and bear responsibility for the submitted content. Courts may reject claims and impose fines or detention for AI-enabled sham litigation while forged evidence may be dealt with under the Civil Procedure Law and criminal responsibility may apply where an offence is established. 

Summary

The Opinions are judicial guidance for courts applying existing laws to AI disputes and are not a new compliance regime for the AI industry. Their practical significance lies in how they connect liability to fault, control, transparency, preventive measures and evidence while supplying scenario-specific rules for personal information, generative content, copyright, open source, patents, products, consumers, automated vehicles and court filings. 

Foreign businesses conducting or planning AI-related activities in China should consider their potential litigation exposure and ensure that relevant technical and training-data records, as well as evidence-preservation procedures, are maintained.

The original publication can be found here (Chinese only).

For more information on China’s AI regulations, contact your CMS client partner or the CMS experts who wrote this article.