China — SPC Opinions on Lawfully Adjudicating AI-Related Dispute Cases (法发〔2026〕10号)

The Opinions on Lawfully Adjudicating Artificial-Intelligence-Related Dispute Cases are guidance issued by China's Supreme People's Court on September 7, 2026 directing all levels of the court system on how to decide AI-related civil, intellectual-property and criminal matters. The court describes it as the first judicial adjudication-rules document on artificial intelligence issued by China's highest adjudicative body (Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10)).

It is not legislation. China has not enacted a dedicated AI statute, and the Opinions expressly proceed by applying existing law — the 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. In that respect it occupies a different layer of the Chinese AI regulatory stack from the CAC-led administrative instruments: the generative AI interim measures, the deep synthesis provisions and the algorithmic recommendation provisions bind providers ex ante, while the Opinions tell courts how to resolve disputes after the fact.

Scope and structure

Part one (articles 1–2) states the guiding ideology and three basic principles: people-centered adjudication, support for innovation, and a firm safety baseline. The third principle directs courts to distinguish general-purpose from special-purpose models and open- from closed-source models by their differences in technical principle, risk spillover and control capability when allocating responsibility among developers, providers and users.

Part two (articles 3–11) covers tort liability, part three (articles 12–16) intellectual property, part four (articles 17–20) procedure, evidence and criminal punishment, and part five (articles 21–24) dispute prevention, adjudication supervision, inter-agency coordination and international judicial exchange (Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10)).

Liability rules

The default rule is fault liability under Civil Code article 1165(1) wherever no statute prescribes strict liability or presumed fault. Fault is assessed on the application scenario, the degree of autonomy, technical and informational transparency, potential risk and its scope, the preventive measures developers and providers took and what was technically possible, and the user's ability to foresee and control the harm (art. 3).

Named practices carry specific rules. Generating and publishing an identifiable virtual likeness without consent supports claims on name and portrait rights, and using a person's voice as training corpus without consent to synthesize an identifiable voice supports a claim on voice interests; making or using a deceased person's virtual likeness to the injury of their name, portrait or reputation gives close relatives a claim under Civil Code article 994 (art. 4). AI-assisted doxxing is invasion of privacy (art. 5). Algorithmic price discrimination attracts tort liability, and celebrity-impersonation selling that amounts to fraud attracts punitive damages under article 55 of the Consumer Rights Protection Law (art. 10). Accidents involving autonomous and assisted-driving vehicles are resolved under the Civil Code and Road Traffic Safety Law, with courts empowered to compel the data-controlling party to produce event-record data (art. 11) (Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10)).

Generative-AI providers face a notice-and-takedown standard: a provider is liable where, after notice with prima facie evidence and the rights-holder's identity, it fails to take necessary measures such as stopping generation of the infringing content. A user who maliciously induces infringing output through prompts is liable in their own right (art. 7). Courts may issue personality-right injunctions ordering a provider to stop supplying a service where infringement is under way or imminent and irreparable harm would follow (art. 8) (Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10)).

Training data and personal information

Article 6 addresses model training directly. Processing personal information that the individual has themselves made public, or that has otherwise been lawfully made public, within a reasonable scope and absent express refusal, is generally not an infringement of personal-information rights; consent is required where the effect on individual rights is major. Reasonableness turns on the necessity and appropriateness of the purpose relative to the model's function, the type and sensitivity of the information, and the context in which it was made public together with the use that could reasonably be expected (Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10)).

Intellectual property

Liability for infringement by AI-generated content is allocated among developer, provider and user on service type, industry characteristics, training-data source, degree of participation, measures taken and profits obtained. A developer asserting non-infringement is to be ordered to produce its training-data source, training-process records, model operating mode and scientific basis (art. 12).

Article 13 provides an appropriate liability exemption for open-source developers and providers, weighed on licence type, the specific limitations of rights, compliance and security measures, and the degree of disclosure. Where a developer supplies code modules free and open-source and publicly explains their function and security risks, a court may find the developer not liable for infringement arising from another's use — the clearest statement in a major jurisdiction's judicial guidance on open-source AI liability. See Open-Weight Frontier Models.

An AI-related invention using technical means that follow natural laws, solving a technical problem and achieving a corresponding effect, is patentable subject matter, excluding inventions to which no natural person made a substantive contribution; where a natural person using AI made a creative contribution to the substantive features, that person is the inventor (art. 14). Data lawfully acquired is protected as a compilation work, as a trade secret, or under article 13 of the Anti-Unfair Competition Law as the case admits, and adversarial-example attacks, fabricated interference data and malicious data labelling that damage AI operating security attract liability (art. 16) (Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10)).

Evidence and fabricated material

Article 18 sets differentiated review standards by evidence type, including specific scrutiny of AI-generated content offered as evidence: the design of the prompts and their effect on the output, similarity to the asserted work, consistency across repeated tests, and the model's training, algorithm design and output filtering.

Article 19 directs dismissal, and fines or detention, where a party fabricates case facts using evidence obtained by deleting or altering generated-content labels, by specific instruction input, selective presentation of results or adversarial interference. It also requires litigation participants who use AI to prepare filings or case-search reports to verify the authenticity of the cited law, interpretations and cases, to disclose the AI assistance on submission, and to bear responsibility for accuracy — a rule addressed to the fabricated-citation problem that courts in several jurisdictions have encountered. See Sycophancy and Hallucination.

What the Opinions leave open

The drafting note records that questions on which consensus is presently difficult were deliberately left blank (留白处理), to be settled later once experience accumulates. Whether AI-generated content attracts copyright protection, and whether training on copyrighted works without permission constitutes infringement, are the two questions in that category (Source: scmp.com; Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10)). Article 6 nonetheless resolves a related question on the personal-information side, and article 12's evidentiary burden on developers bears on training-data disputes without settling the underlying copyright rule.

Relationships

• depends-on: Supreme People's Court Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10) — the primary text
• related: China — Interim Measures for the Management of Generative AI Services, China — Provisions on the Administration of Deep Synthesis Internet Information Services, China — Internet Information Service Algorithmic Recommendation Management Provisions, China — Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents, China — Interim Measures for the Administration of AI Anthropomorphic Interaction Services — the administrative layer of the same regulatory stack
• related: Cyberspace Administration of China (CAC) — the regulator named in the article 23 coordination duty
• related: AI Liability, AI and Tort Liability — the liability allocation questions
• related: AI Copyright — the questions expressly deferred
• related: Synthetic Media / Deepfakes — the article 4 likeness and voice provisions
• related: EU vs. US AI Regulation: A Deep Comparison — comparative regulatory approaches