The Opinions on Lawfully Adjudicating Artificial-Intelligence-Related Dispute Cases (最高人民法院关于依法审理涉人工智能纠纷案件的意见, issuing number 法发〔2026〕10号) were released by China's Supreme People's Court on September 7, 2026 at a press conference in Beijing. The court describes the document as the first judicial adjudication-rules instrument on artificial intelligence issued by the country's highest adjudicative body. It runs to five parts and 24 articles.
The Opinions are guidance to lower courts on applying existing law, not new legislation. The drafting note states that China has not enacted a dedicated AI statute (目前我国尚未出台专门的人工智能法) and that the document therefore proceeds on the principle of making full use of existing provisions, citing 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 as its legal bases.
The full Chinese text is held as a raw source; no translation has been made, and quotations below render the Chinese with an English gloss.
Issuance and framing
The Opinions were released by Vice-President Tao Kaiyuan (陶凯元), with Research Office Director Zhou Jiahai (周加海), No. 3 Civil Division Presiding Judge Li Jian (李剑) and Research Office Deputy Director Si Yanli (司艳丽) taking questions; the conference was chaired by court spokesperson Ji Zhongbiao (姬忠彪).
The drafting note situates the document within the party-state's AI policy program, citing Xi Jinping's April 2025 remarks at the twentieth Politburo collective study session on ensuring AI is "safe, reliable, controllable" (安全、可靠、可控), the Fifteenth Five-Year Plan outline's direction to explore rules for ownership of AI-generated outputs and for allocating rights and responsibilities among developers, operators and users, and the State Council's Fifteenth Five-Year Plan for intellectual property protection. The stated aim is to draw legal "red lines" while supporting innovation — in the note's metaphor, letting the thousand-li horse of AI both run fast and run steadily (既跑得快，又跑得稳).
The document records an explicit decision to leave contested questions unresolved: 对于目前难以形成共识的问题，《意见》做了留白处理 — questions on which consensus is presently difficult are left blank, to be settled later by an appropriate method once experience accumulates. Whether AI-generated content attracts copyright, and whether training on copyrighted works without permission infringes, fall in that gap.
Part one: guiding thought and principles (articles 1–2)
Article 1 sets the guiding ideology. Article 2 states three basic principles: people-centered (以人为本), directing courts to strengthen protection of livelihood interests in the digital era, guard against and eliminate algorithmic discrimination, and secure individuals' rights to know and to choose in automated decision-making; supporting innovation and development (支持创新发展), calling for a tolerant and prudent balance between rights protection and industrial development, equal treatment of developers, and particular attention to the interests of small and medium-sized enterprises; and building a firm safety baseline (筑牢安全底线), which directs courts to distinguish between general-purpose and special-purpose models and between open- and closed-source models according to their differences in technical principle, risk spillover and control capability when allocating responsibility among developers, providers and users.
Part two: tort liability (articles 3–11)
Fault as the default (art. 3). Where no statute prescribes strict liability or presumed fault, courts apply the ordinary fault principle of Civil Code article 1165(1). In assessing fault and its degree, courts are directed to weigh the application scenario, the degree of autonomy, technical and informational transparency, the potential risk and its scope, the measures developers and providers took to prevent or reduce harm and what was technically possible, and the user's ability to foresee and control the harm.
Face-swapping, voice imitation and digital resurrection (art. 4). Generating and using or publishing an identifiable virtual likeness of a natural person without consent supports a claim for infringement of name and portrait rights. Using a natural person's voice as training corpus without consent to synthesize a voice identifiable as theirs — imitating timbre, intonation and pronunciation style — supports a claim on voice interests. Manipulating a generated likeness or voice to perform improper acts or make false statements that lower social evaluation constitutes defamation. Where a deceased person's virtual likeness is made or used without authorization to the injury of their name, portrait or reputation, close relatives may claim under Civil Code article 994.
AI-assisted doxxing (art. 5). Using AI to track and analyze a person's phone number, online accounts and social-media presence for the purpose of prying into private information, and then obtaining, leaking, publishing or using it to disturb private life, constitutes invasion of privacy — the practice the document names 网络开盒 and 人肉搜索.
Publicly available personal information as training corpus (art. 6). Processing, within a reasonable scope and for model training, personal information an individual has themselves made public or that has otherwise been lawfully made public, absent express refusal, is generally not an infringement. Consent is required where the effect on individual rights is major. Reasonableness is assessed on the necessity and appropriateness of the purpose relative to the model's function, the type and sensitivity of the information and its potential effect, and the context in which the individual made it public together with the use they could reasonably expect.
Generative-AI provider liability (art. 7). A provider whose system automatically generates content infringing reputation or privacy, and who fails to take necessary measures such as stopping generation after notice from the rights-holder, bears liability for the resulting harm; the notice must include prima facie evidence and the rights-holder's true identity. A user who maliciously induces infringing output through prompts bears liability; where the provider then fails to stop generation or block the relevant instructions after notice, both may be held liable under Civil Code article 1195.
Personality-right injunctions (art. 8). Courts may issue an injunction ordering a person to stop conduct, or ordering a network or generative-AI service provider to stop providing the service, where an applicant shows AI-enabled infringement of personality rights is under way or imminent and would cause irreparable harm; the measure must not exceed what is necessary.
AI product liability (art. 9). Courts apply the Product Quality Law's definition of "product" to AI products embodied in physical form. In assessing whether an unreasonable danger exists, they weigh the product's nature and use, its autonomous learning capability, its upgrade history, the degree of user control, and conformity with national and industry standards, with particular scrutiny of whether the producer or seller gave truthful explanation and clear warning of the applicable scenarios, inherent limitations and foreseeable risks.
Algorithmic price discrimination and celebrity-impersonation selling (art. 10). Unreasonable differential treatment in price or other trading conditions for the same goods or service — the practice known as 大数据杀熟 — attracts tort liability. Assessment turns on whether the treatment substantially restricts or harms the consumer's rights to know, to choose freely and to fair dealing, whether the terms were individualized from consumption preferences, willingness or ability to pay and browsing history, and whether it violates good faith and commercial ethics. Where AI-enabled 仿冒名人带货 (impersonating a celebrity to sell goods) amounts to fraud, punitive damages under article 55 of the Consumer Rights Protection Law are available.
Driving-automation accidents (art. 11). Liability for road accidents involving autonomous and assisted-driving vehicles rests on the Civil Code and the Road Traffic Safety Law. Where a product defect caused the accident, producers or sellers are liable; where a defect in an assisted-driving vehicle combined with driver fault produced the same harm, both the driver and the producer or seller may be held liable. False or misleading advertising of a vehicle's automation level, intelligence, performance or intended use supports consumer claims. To establish the cause of an accident, courts may require the producer, seller or operator — as the party controlling the data — to supply true and complete event-record data within the necessary scope.
Part three: intellectual property (articles 12–16)
Infringement by AI-generated content (art. 12). Responsibility among developer, provider and user is allocated on the service type, industry characteristics, the source of training data, each party's degree of participation, the measures taken and the profits obtained. A developer asserting non-infringement is to be ordered to produce the training-data source, training-process records, model operating mode and scientific basis in support. A user who knew or should have known of a prior work and generated a substantially similar work without a reasonable defence is liable.
Open-source liability exemption (art. 13). Courts allocating liability among open-source developers, providers and downstream developers weigh the licence type, the specific limitations of rights, compliance and security measures, and the degree of information disclosure, and are to grant open-source developers and providers an appropriate exemption from liability (依法给予开源软件开发者、提供者适当的责任豁免). Where a developer supplies code modules free and open-source for AI development and publicly explains their function and security risks, a court may find the developer not liable for infringement caused by another's use.
Patentability and inventorship (art. 14). An AI-related invention that uses technical means following natural laws, solves a technical problem and achieves an effect consistent with natural laws is patentable subject matter — excluding inventions that violate law or social morality, harm the public interest, or to which no natural person made a substantive contribution. Where a natural person using AI made a creative contribution to the substantive features of an invention, that person is the inventor.
Technology contracts (art. 15). In development, transfer, licensing, consulting and service contract disputes, breach is assessed on the contract terms, the characteristics of AI research and development, and whether the developing party made reasonable efforts.
Data rights (art. 16). Data lawfully acquired through collection, derivation, assignment or licence, and in which the developer holds corresponding interests, is protected. Data and datasets that constitute compilations or otherwise meet the requirements of a work are protected under the Copyright Law; those constituting trade secrets under the Anti-Unfair Competition Law; and those doing neither may still ground liability under article 13 of that law. Monopoly agreements or abuse of dominance achieved through data and algorithms attract liability, as do technical means that damage the operating security of AI systems, including fabricated interference data, malicious data labelling and adversarial-example attacks.
Part four: procedure, evidence and crime (articles 17–20)
Article 17 directs courts to guide the parties' evidence-gathering, to collect evidence on application or of their own motion where a party cannot obtain it, to review preservation applications promptly, and to draw an adverse inference where a party controlling documentary or electronic evidence refuses without good reason to produce it. It directs use of people's assessors, appraisers, expert assistants and technical investigation officers on questions of technical principle and operating mechanism.
Article 18 sets differentiated review standards by evidence type: for big-data analysis reports, the source of the raw data, the cleaning rules and the scientific soundness of the analytical method; for blockchain-preserved evidence, the authenticity of the data before it was written to the chain and the reliability of the platform. Where a party offers AI-generated content as evidence of infringement, courts weigh the design of the prompts and their effect on the output, the similarity between output and the asserted work, consistency across repeated tests, and the model's training, algorithm design and output-filtering mechanisms.
Article 19 addresses fabricated evidence. A party who obtains false evidence by deleting or altering generated-content labels, by specific instruction input, selective presentation of results or adversarial interference, and who fabricates the basic facts of a civil case to bring false litigation, is to have the claim dismissed and may be fined or detained, with criminal liability where the conduct amounts to a crime. It also requires that participants who use AI to produce court filings or case-search reports verify the authenticity and accuracy of the cited law, judicial interpretations and cases before submission, disclose the AI assistance when filing, and bear responsibility for accuracy.
Article 20 addresses criminal liability for AI-enabled fraud, insult, defamation, damage to commercial reputation, infringement of citizens' personal information, unlawful acquisition of computer-system data, and the production, sale and dissemination of obscene material. It separately provides for criminal liability where a driver activates an assisted-driving function and uses privately installed accessories to evade the system's monitoring, causing an accident.
Part five: institutional mechanisms (articles 21–24)
Article 21 directs use of diversified dispute-resolution mechanisms with industry regulators, specialist mediation bodies and academic experts. Article 22 provides for elevated jurisdiction (提级管辖) in cases involving major interests, difficult or novel questions, or the establishment of rules, and for use of the courts' case database for guidance. Article 23 directs coordination with the cyberspace, public-security, procuratorial and market-regulation authorities toward what the text calls a full-chain combination of industry self-regulation, administrative supervision and judicial protection. Article 24 addresses cross-border AI and data disputes and international judicial exchange.
Reception
The South China Morning Post reported the release on September 8, 2026 as China's highest court setting out AI red lines on deepfakes and privacy, and noted that the guidance did not resolve whether AI-generated content qualifies for copyright protection or whether training on copyrighted works without permission infringes (Source: scmp.com). The court's own drafting note frames that omission as deliberate blank-leaving rather than oversight.
Provenance
Captured in full from court.gov.cn, which the court states is its only official identity on the internet. The page carries its own source line (最高人民法院新闻局) and timestamp of 2026-09-07 19:22:29; the document carries the issuing number 法发〔2026〕10号 and is signed 最高人民法院, 2026年9月7日. The article count is corroborated internally: the drafting note states five parts and 24 articles, and the captured text runs from article 1 to article 24 without gaps. A companion question-and-answer page published by the court the same day is a further primary document and is not held here.
Relationships
- related: China — SPC Opinions on Lawfully Adjudicating AI-Related Dispute Cases (法发〔2026〕10号) — the tracking page for this instrument
- depends-on: 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 — the administrative layer this judicial guidance sits alongside
- related: China — Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents, China — Interim Measures for the Administration of AI Anthropomorphic Interaction Services — contemporaneous Chinese AI instruments
- related: AI Liability, AI and Tort Liability — the liability questions the Opinions allocate
- related: Synthetic Media / Deepfakes — the face-swapping, voice-imitation and digital-resurrection provisions
- related: AI Copyright — the questions the Opinions expressly leave open
- related: EU vs. US AI Regulation: A Deep Comparison — comparative regulatory approaches
- related: Cyberspace Administration of China (CAC) — the administrative regulator with which article 23 directs coordination