On 7 September 2026 the Supreme People’s Court issued its Opinion on the Trial of Cases Involving Artificial Intelligence Disputes (Fa Fa [2026] No. 10), in five parts and twenty-four articles. This is the first occasion on which the highest judicial organ has addressed, in a dedicated document, questions including AI face swapping and voice synthesis, the “resurrection” of deceased persons, liability for AI hallucination, the use of personal information in training data, liability for open-source software, and copyright in AI-generated content. The Opinion creates no new law: it applies existing rules — the Civil Code, the Copyright Law, the Personal Information Protection Law and others — to AI scenarios, and sets out the reasoning by which courts at every level should decide such cases. For companies building AI products, the Opinion draws the boundaries of liability in product design and operation in advance.
The twenty-four articles fall into three blocks: tortious liability (Articles 3 to 11), intellectual property (Articles 12 to 16), and evidence and allocation of liability in litigation (Articles 17 to 20). The most direct effect on companies is that three obligations — content labelling, notice and takedown, and the provenance of training data — are made concrete.
What the Opinion is
The Opinion bears the reference Fa Fa [2026] No. 10 and is a document of judicial policy and trial guidance issued by the Supreme People’s Court, not a judicial interpretation. China has not yet enacted a dedicated AI statute, and the Opinion therefore rests on existing law — the Civil Code, the Cybersecurity Law, the Data Security Law, the Copyright Law, the Anti-Unfair Competition Law, the Consumer Rights Protection Law, the Personal Information Protection Law and the Civil Procedure Law — in offering guidance on concepts and rules for AI-related disputes. Its practical weight lies in its direct guidance to courts at every level, which gives it greater authority than academic commentary.
Articles 1 and 2 establish three basic principles: a people-centred approach, strengthening the protection of livelihood interests in the digital age, preventing and eliminating algorithmic discrimination and safeguarding the individual’s right to be informed and to choose in automated decision-making; support for innovation, balancing protection of rights with industrial development on an inclusive and prudent basis and strengthening equal protection for small and medium-sized enterprises; and a firm safety baseline, distinguishing general from specialised models and open-source from closed-source models, and allocating responsibility reasonably among developers, providers and users.
Tortious liability (Articles 3 to 11)
Basis of liability (Article 3). Where the law does not expressly provide for strict liability or liability on a presumption of fault, liability is determined on the basis of fault under Article 1165(1) of the Civil Code. In assessing whether fault exists and its degree, the court considers the application scenario, the degree of autonomy, the transparency of technology and information, the potential risk and its reach, the measures taken by the developer or provider to prevent infringement and the technical feasibility of further measures, and the user’s capacity to foresee and control the harm. The practical meaning of this article is that compliance steps taken by a company may serve as a defence on fault: safety measures, moderation records and risk warnings must all be documented, not merely processed internally.
AI face swapping, voice synthesis and the resurrection of the deceased (Article 4). Without the consent of a natural person, the use of AI to process that person’s name or likeness so as to generate an identifiable virtual digital image, and its use or publication, infringes personality rights including the rights to name and likeness; the use of a natural person’s voice as training material without consent, imitating timbre, intonation and pronunciation to generate synthetic speech identifiable as that person, infringes voice rights; the manipulation of an identifiable person’s virtual image or voice to commit improper acts, make false statements or lower the person’s standing infringes reputation; and the unauthorised production or use of a deceased person’s virtual digital image, causing infringement of name, likeness or reputation, entitles close relatives to claim under Article 994 of the Civil Code. For products involving the likeness or voice of real people, the absence of authorisation is no longer a grey area but a clear point of infringement risk, and a chain of authorisation and a review of source material are required.
Doxxing (Article 5). The use of AI, for the purpose of prying into private matters, to track and analyse publicly available information such as telephone numbers, online accounts and social media of a specific person, to obtain private information, or to leak or publish the private information obtained, or to use the information obtained to disturb the peace of private life, infringes privacy. The use of AI to photograph, peer at or eavesdrop on private spaces or private activities equally constitutes infringement, save where otherwise provided by law or expressly consented to by the person.
Training data and publicly available personal information (Article 6). Processing, within a reasonable scope and for model training, personal information that the individual has made public or that is otherwise lawfully public, where the individual has not expressly refused, is generally not treated as infringement of personal information rights; but where the processing has a significant effect on the individual’s rights, consent must be obtained in accordance with law. In determining a reasonable scope, the court considers the necessity and appropriateness of the purpose and model function, the type, sensitivity and potential impact of the personal information, and the context in which it was made public and the scope of use reasonably to be expected. This article leaves room for training data, but it is not a blanket authorisation: companies need records of the provenance of public data, sensitivity grading and a mechanism for responding to express refusals, and a separate consent route for processing likely to have a significant effect.
Liability of generative AI service providers (Article 7). A generative AI service provider is subject to the notice-and-takedown logic: where content automatically generated by AI infringes personality rights such as reputation or privacy, and the provider fails, after notice from the rights holder, to take necessary measures such as ceasing to generate the infringing content, the provider is liable for the resulting harm. The notice must include prima facie evidence of infringement and reliable identification of the rights holder. Where a network user maliciously induces the generation of infringing content by entering infringing prompts and causes harm, that user is liable in accordance with law; and where the provider, having received notice, fails promptly to cease generation and to block the relevant instructions, it may be held liable together with the user under Article 1195 of the Civil Code. For product operation, a complaints channel, a moderation time limit and the capacity to cease generation and block instructions are therefore baseline requirements.
Injunctions to restrain infringement of personality rights (Article 8). Where evidence shows that a person is infringing, or is about to infringe, personality rights by means of AI, and that failure to restrain the conduct promptly would cause harm difficult to remedy, the court may grant an injunction restraining the conduct, or restraining a network service provider or generative AI service provider from providing the relevant service. The court adopts measures no more extensive than necessary, having regard to the type of right infringed, the manner of the conduct and the scope and degree of the harm.
Product liability for AI products (Article 9). Under the definition of a product in the Product Quality Law, an AI product embodied in a physical medium is subject to the product liability rules. In determining whether there is an unreasonable danger to personal or property safety, the court considers the nature and purpose of the product, its capacity for autonomous learning, its upgrades and updates, the degree of user control over the system, and conformity with national and industry standards, and examines in particular whether the producer or seller has given truthful explanations and clear warnings as to the applicable scenarios, inherent limitations and foreseeable risks.
Algorithmic price discrimination and impersonation of celebrities in live commerce (Article 10). The use of algorithms to impose unreasonable differential treatment in transaction price or other terms, causing harm, gives rise to liability in tort. In determining liability the court considers whether the consumer’s right to be informed, to choose freely and to trade fairly has been materially restricted, whether personalised transaction terms have been formed on the basis of consumption preference, willingness or ability to pay or browsing history, and whether the reason for the difference is proper, adequate and non-discriminatory. Where AI is used to impersonate a celebrity in live commerce and the conduct amounts to fraud, the consumer may claim punitive damages.
** Autonomous and assisted driving (Article 11).** Where an accident is caused by a defect in the vehicle product, the producer and seller are liable; where a defect in an assisted driving function combines with the driver’s fault to cause the same harm, claims may be brought against both the driver and the producer or seller. False or misleading promotion of the level of automation, intelligence, performance or purpose gives rise to civil liability. To establish the cause of an accident, the court may require the party controlling the data to provide, within the necessary scope, a true and complete record of the autonomous or assisted driving event.
Intellectual property (Articles 12 to 16)
Infringement of copyright by AI-generated content (Article 12). In determining whether AI-generated content infringes copyright, the court considers the type of service, the characteristics of the industry, the provenance of training data, the degree of participation of each party, the measures taken and the profits obtained, and allocates liability reasonably among developer, provider and user. Where a developer raises a defence of non-infringement, it should be ordered to provide the provenance of the training data, records of the training process, the mode of operation of the model and the scientific and theoretical basis. A user who knows or ought to know of the existence of a prior work, and who generates a work substantially similar to it without a reasonable defence, is liable for infringement.
Exemption from liability for open-source software (Article 13). In cases concerning open-source software, in determining the liability of developers and providers, including subsequent developers and providers, the court considers the type of open-source licence, the content of the restrictions, the safety and compliance measures taken and the degree of disclosure, and grants an appropriate exemption. Where a party provides, free of charge and on an open-source basis, code modules required for the development of AI software, and publicly explains their functions and safety risks, that party may be held not liable for infringement resulting from another’s use of the module. For the many companies that build on open-source models, this article is clearly favourable, but the exemption is conditioned on public explanation of function and risk, and open-source compliance documentation must be kept in step.
Patent grant and validity for AI inventions (Article 14). An AI invention that employs technical means following the laws of nature, solves a technical problem and achieves a technical effect conforming to the laws of nature is to be recognised as patentable subject matter, save where it contravenes the law, social morality or the public interest, or where no natural person has made a substantive contribution. Where a natural person has made a creative contribution to the substantive features of an invention completed with the aid of AI, that person is to be recognised as the inventor. Where the specification is expressed to a degree that a person skilled in the art can carry out the invention, the sufficiency of disclosure requirement is met.
Breach of technology contracts (Article 15). In disputes over contracts for the development, transfer, licensing, consultancy or provision of AI technology, liability for breach is determined in accordance with the contract, having full regard to the characteristics of AI research and development and to whether the developer has exercised reasonable efforts.
Use of data and data rights (Article 16). Data lawfully obtained by collection and generation, derivative creation, assignment or licence, and the rights enjoyed in it, are protected. Data or datasets constituting a compilation are protected under the Copyright Law; those constituting a trade secret are protected under the Anti-Unfair Competition Law; and where the data do not constitute a trade secret but the impugned conduct violates Article 13 of the Anti-Unfair Competition Law, liability follows. The use of data and algorithms to conclude monopoly agreements or to abuse a dominant position, and the adoption of technical means such as fabricated or interfering data, malicious labelling or adversarial sample attacks to impair the safety of AI operation, also give rise to liability.
Evidence and liability in litigation (Articles 17 to 20)
Establishment of facts and proof (Article 17). The court should give active guidance in the proceedings, leading the parties to adduce evidence promptly and comprehensively on the disputed facts, and may where necessary collect evidence of its own motion. Where a party controlling documentary or electronic evidence refuses to produce it without proper reason, the court may accept the other party’s assertion that the content of the evidence is adverse to the controlling party. On professional questions concerning the principles and operation of AI technology, the court may rely on people’s assessors, expert witnesses, expert assistants and technical investigators.
Examination of evidence (Article 18). The authenticity and integrity of the generation, collection, storage and transmission of electronic data are examined with particular care; for evidence in the form of big-data analysis reports, the provenance of the underlying data, the cleaning rules and the soundness of the analytical method are examined; and for blockchain-stored evidence, the authenticity of the data before it was recorded on the chain and the reliability of the technical platform are examined. Where a party relies on AI-generated content as evidence of infringement, the court considers the design of the prompts and their influence, the degree of similarity between the generated content and the work in respect of which rights are claimed, the consistency of repeated testing, and the model training, algorithm design and content filtering mechanisms.
Improper obtaining of evidence and false litigation (Article 19). Where false evidence is obtained by deleting or altering the labels on generated synthetic content, by particular instruction inputs, by selective presentation of results or by adversarial interference, and a false claim is brought on a fabricated factual basis, the court should dismiss the claim and may impose a fine or detention, with criminal liability where an offence is constituted. Where a document or case-law research report submitted by a participant in the proceedings is AI-generated, the participant should verify the authenticity and accuracy of the statutes, judicial interpretations and cases before submitting it to the court, explain the use of AI assistance and take responsibility for the truth of the content.
Commission of offences by means of AI (Article 20). The use of AI to commit fraud, insult, defamation, harm to business reputation or product reputation, infringement of citizens’ personal information, unlawful acquisition of computer information system data, or the production, sale or dissemination of obscene material, constitutes an offence and gives rise to criminal liability in accordance with law.
What companies should prepare
The Opinion does not supply a simple answer as to whether a given product may be built. It makes the boundaries of liability in AI scenarios concrete and provable. In day-to-day operations, four categories of material should be prepared in advance: the chain of authorisation (authorisations and source material for the likeness and voice of real people); the provenance of data (records of the source of training data and sensitivity grading); records of handling (the receipt, moderation and disposal of infringement notices, with a full audit trail); and the language of promotion (truthful explanation and warning as to applicable scenarios, inherent limitations and foreseeable risks). These four categories serve both as compliance requirements and as evidentiary assets in litigation.
Several points are readily misread and should be stated separately. The Opinion is not a judicial interpretation, but it directly guides the reasoning of courts at every level and carries strong persuasive authority in practice. The use of publicly available online data to train a model is not necessarily infringing: Article 6 leaves room within a reasonable scope, and the essential questions are the proof of that scope and the grading of sensitive information. A user of open-source software cannot treat an upstream exemption as its own: the exemption in Article 13 is directed to the open-source developer and provider, and the user remains responsible for its own conduct. Where a user generates infringing content with a product, the provider’s liability begins upon receipt of a valid notice, and the capacity to handle complaints is therefore essential infrastructure. AI-generated content may be adduced as evidence, but it is examined more strictly, and Article 19 further requires the submitting party to take responsibility for its truth and to explain the use of AI.
The common direction of these rules is to move compliance capacity forward into product design, rather than to assemble it after a notice or an injunction has been received.
For assistance with personality-right and copyright risk in AI products, training data compliance, mechanisms for handling infringement notices, intellectual property strategy and protection, or criminal compliance and regulatory response, please contact Geason Law Offices. Further reading: intellectual property protection for AI companies and the AI data compliance risk checklist.
This article is prepared on the basis of the Opinion of the Supreme People’s Court on the Trial of Cases Involving Artificial Intelligence Disputes (Fa Fa [2026] No. 10, issued 7 September 2026). It is for reference only and does not constitute legal advice. Individual matters should be assessed on their own facts and evidence.