China AI Dispute Adjudication Rules

Overview

The Supreme People's Court of the People's Republic of China issued the Opinions on Adjudicating Artificial Intelligence-Related Dispute Cases in Accordance with Law, 法发〔2026〕10号, on 7 September 2026. This judicial guidance directs people's courts at all levels to apply existing substantive and procedural law to AI disputes. It is classified here as a Regulation because of its mandatory court-facing rules, not as non-binding recommendations; it is not a primary AI statute. The official publication containing the complete instrument supplies the document number, signature date, and release information.

The five parts contain 24 clauses: general requirements, clauses 1–2; tort disputes, clauses 3–11; intellectual property disputes, clauses 12–16; procedural rules and unlawful conduct, clauses 17–20; and judicial guidance and working mechanisms, clauses 21–24. The three basic principles are putting people first, supporting innovative development, and maintaining a safety baseline. Courts must balance civil-rights protection with technological and industrial innovation, considering application scenarios, risk magnitude, technical differences, and developers', providers', and users' control. The Opinions address identifiable harms and litigation problems, not a comprehensive AI licensing system.

Definitions

The Opinions contain no general definitions article or universal statutory definition of artificial intelligence. Clause 2 distinguishes general-purpose and specialized models, and open-source and closed-source models, according to technical principles, risk spillovers, and control capabilities when allocating responsibility. Clause 9 identifies physically embodied AI products through the Product Quality Law's definition of a product. Defect analysis considers unreasonable dangers to personal or property safety, product nature and use, autonomous learning, updates, user control, applicable standards, and truthful explanations and clear warnings about limitations and foreseeable risks.

Other concepts operate through contextual legal tests. Clause 4 concerns digital images and synthetic voices capable of identifying a particular person. Clause 6 assesses the reasonable scope of processing publicly disclosed personal information through purpose, necessity and appropriateness for model functions, information type and sensitivity, potential effects, disclosure context, and reasonably expected uses. Clause 14 assesses AI-related inventions through technical means, problems, and effects consistent with natural laws, together with substantive human contribution. Clause 18 examines generated evidence through prompts, similarity, consistency across repeated tests, training, algorithms, and filtering mechanisms.

Governance and Institutional Framework

The Supreme People's Court provides national judicial direction; people's courts apply the rules in individual cases. Clause 22 strengthens supervisory adjudication and guidance through higher-level assumption of jurisdiction. Cases involving major interests, difficult and complex new categories, rule-establishing significance, or a need to unify legal application standards are to be heard by higher-level courts. Courts must fully use the People's Courts Case Database for case guidance and consistency. This adjudicatory framework does not create a separate AI regulator, certification authority, or administrative approval process.

Clauses 21 and 23 require coordinated dispute prevention and resolution with AI industry authorities, specialized industry mediation organizations, experts, and scholars, including stronger preliminary mediation. Courts must improve cooperation with cyberspace authorities, public security organs, procuratorial organs, and market regulators through judicial recommendations and cooperation mechanisms. Enterprises, platforms, and industry associations should be guided toward orderly self-governance. The framework combines industry self-discipline, administrative supervision, and judicial protection. Public hearings and explanations of law through cases clarify rights, duties, and lawful behavioral boundaries.

Key Focus Areas

Clauses 4–8 protect personality rights, privacy, and personal information. Subject to legal exceptions, courts support claims involving unauthorized identifiable digital likenesses and synthetic voices trained on a person's voice. Manipulated images or voices that diminish reputation can infringe reputation rights; close relatives may protect deceased persons' interests under Civil Code Article 994. Using AI to track public information to uncover or disclose private information, or disturb private life, can infringe privacy. Public disclosure does not authorize unrestricted training: processing must remain within reasonable scope, respect express refusal, and obtain consent where personal rights are significantly affected.

Clauses 9–11 address physical AI product defects, consumer treatment, and vehicle accidents. Courts examine misleading driving-automation claims and may require necessary, truthful, complete accident-event records from data controllers. Algorithmic differential treatment is assessed against informed choice, independent choice, fair dealing, justification, and nondiscrimination. Fraudulent AI celebrity impersonation can support punitive compensation under Consumer Rights Protection Law Article 55.

Clauses 12–16 address copyright, open-source responsibility, patents, technology contracts, and lawful data interests. Copyright responsibility depends on the service, data sources, participation, precautions, and benefits. Open-source exemptions are qualified, not automatic. Patent analysis requires relevant human contribution and adequate disclosure. Technology-contract decisions account for agreed terms, research characteristics, and reasonable efforts. Lawful data interests receive protection through applicable copyright, trade-secret, and competition rules, including against malicious data manipulation or adversarial attacks harming model operation.

Implementation Framework

Implementation occurs through fact-finding, attribution, remedies, and procedural management. Clause 3 applies fault-based tort liability under Civil Code Article 1165(1), unless legislation expressly requires no-fault liability or presumed fault. Courts assess application context, autonomy, transparency, potential risks, preventive measures, technical feasibility, and users' ability to foresee and control harm. Clause 7 requires timely necessary responses to qualifying personality-rights infringement notices containing preliminary evidence and the rights holder's real identity information. Responses include stopping infringing generation and, for malicious user prompting, blocking relevant instructions. Timeliness is required, but no numerical response deadline is established.

Clauses 12 and 17 govern evidentiary implementation. Developers asserting non-infringement must substantiate their position with training-data sources, training records, model operating modes, and scientific theoretical grounds. Courts must guide timely and complete proof, investigate evidence where appropriate, and promptly examine applications to preserve key technical evidence. If a party unjustifiably refuses to produce documentary or electronic evidence under its control, the court may accept the opposing party's assertion that it is unfavorable to its controller. People's assessors, appraisers, expert assistants, and technical investigation officers may assist with specialized questions. These are litigation mechanisms, not universal predeployment documentation or audit requirements.

Monitoring and Evaluation

Oversight relies on judicial supervision, case guidance, and case-specific technical scrutiny rather than periodic compliance reporting. Clause 18 requires examination of electronic data authenticity and integrity throughout generation, collection, storage, and transmission. Big-data analysis reports require scrutiny of original sources, cleaning rules, and the scientific validity of analytical methods. Blockchain-preserved evidence requires examination of pre-chain data authenticity and platform reliability. AI-generated content offered as infringement evidence must be evaluated for prompt design and other model-related influences rather than treated as conclusive.

Evaluation also scrutinizes risk allocation and conduct. Product cases examine standards, learning capabilities, updates, user control, and warnings; vehicle cases may require complete and truthful event records within the necessary scope. Copyright cases assess provenance, participants' involvement, preventive measures, and gains. Clause 22 supports consistency through higher-level handling and case-database guidance, while clause 23 supports long-term coordinated governance. No annual AI audits, mandatory incident-reporting intervals, quantitative performance indicators, or scheduled review date are prescribed. The oversight model is primarily adjudicatory and institutionally coordinated.

Penalties, Liability, and Appeals

Civil consequences depend on applicable law and case facts, including tort liability, defective-product liability, vehicle-accident compensation, contractual liability, and punitive compensation for fraudulent AI celebrity impersonation. Under clause 8, natural persons, legal persons, and unincorporated organizations may seek personality-rights injunctions where evidence shows ongoing or imminent unlawful infringement and irreparable harm without timely intervention. Courts may order cessation of conduct or relevant services, but measures must stay within necessary limits. Clause 13 permits appropriate open-source liability exemptions after assessing license terms, restrictions, safety compliance, and disclosures; it does not exempt all open-source participants from all infringement claims.

Clause 19 requires dismissal of claims in the specified AI-enabled fabricated-litigation circumstances and permits fines or detention according to seriousness, with criminal responsibility where an offense is constituted. AI evidence forgery obstructing adjudication is addressed under Civil Procedure Law Article 114. Litigation participants must verify AI-generated filings and research materials, explain AI assistance when submitting them, and remain responsible for accuracy and authenticity. Clause 20 addresses AI-enabled crimes and criminally culpable circumvention of assisted-driving monitoring leading to accidents. The Opinions establish no independent fine amounts, sentencing ranges, appeal deadlines, or special AI appeal procedure; existing legal procedures remain applicable.

Relationship to Other Instruments

The preamble invokes 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. Other provisions invoke the Product Quality Law, Road Traffic Safety Law, and Patent Law. The Opinions organize existing law for AI disputes rather than replace it. The official Civil Code text and National Laws and Regulations Database provide statutory context.

Express Civil Code references include Article 1165(1) on fault attribution, Article 994 on deceased persons' interests, Article 1195 on network-related infringement responsibility, Article 1172 on combined causes of harm, and Book Seven, Chapter Four on product liability. Other references include Consumer Rights Protection Law Article 55, Anti-Unfair Competition Law Article 13, and Civil Procedure Law Article 114. Clause 14 addresses patent eligibility, human inventorship, and sufficient disclosure without establishing AI as an inventor. Clause 16 protects lawfully acquired data interests according to the data's legal character and challenged conduct, including compilations, other copyright works, trade secrets, and unfair competition. Clause 15 considers technology-contract terms, research characteristics, and reasonable efforts. No comprehensive copyright-ownership rules for AI outputs or blanket permission to train on copyrighted works are established.

International Alignment

Clause 24 directs courts to handle AI-related and cross-border data disputes appropriately and strengthen international judicial exchange and cooperation based on mutual respect, equality, and mutual benefit. It calls for coordination and mutual learning concerning rules and seeks a fairer, more reasonable, inclusive, and open AI governance system. It emphasizes promoting and protecting diversity in AI ecosystems and human civilization. Alignment is an objective of judicial dialogue, not incorporation of a specified foreign statute, supranational regulation, technical standard, or treaty.

Cross-border disputes remain subject to lawful adjudication alongside domestic rights protection and judicial mechanisms. Clause 24 creates no reciprocal certification, mutual recognition of AI assessments, new transfer authorization, or exemption from domestic data law. It names no partner countries, organizations, implementation dates, or cooperation procedures. Its concrete duty is to advance judicial exchange and rule-related mutual learning while properly resolving AI and cross-border data cases; its broader objective is inclusive governance respecting technological and civilizational diversity.

Implementation Timeline

Sources and References