Introduction

The Supreme People’s Court (‘SPC’) recently issued the Opinion on the Adjudication of AI-Related Disputes in Accordance with Law (the ‘Opinion’), providing further adjudicative guidance on issues such as the determination of tort liability, the protection of rights and interests, and fact-finding arising from AI applications. Grounded in the Civil Code and relevant special statutes, and taking into account the technical features and application scenarios of AI, the Opinion refines the application of existing legal rules. In a given case, what kind of conduct the relevant party has undertaken, what capacity it has to foresee and control the risk, and whether it has adopted reasonable and feasible preventive measures will be important factors in deciding whether it bears liability.

From AI face-swapping and voice imitation to humanoid robots and intelligent driving, AI may infringe personality rights and intellectual property rights through content generation and information processing, and may also endanger life, body, health and property safety in physical operation; at the same time, AI agents possess perception, decision-making and execution capabilities, which expands the range of legal interests that may be infringed and makes the principles of attribution and the responsible parties more diverse. Following this analytical approach, and taking the Opinion as its basis, this article discusses the liability boundaries of generative AI and physical AI from the perspective of the interests infringed, the distinction between products and services, and the actual conduct of the relevant parties.

I. Basis of Liability: Distinguishing Services from Products and Reasonably Determining the Responsibility of the Parties

Article 3 of the Opinion makes clear that where the law provides for no-fault liability or liability based on presumption of fault, those provisions apply; where there is no special provision, the general fault-based principle under Article 1165(1) of the Civil Code applies. For example, where a personal information processor infringes personal information rights and interests and causes damage, the presumption-of-fault rule in Article 69 of the Personal Information Protection Law applies; where damage is caused by a product defect, the producer’s liability does not require fault.

In my view, whether a product is supplied or a service is provided should be an important basis for distinguishing the principles of attribution: AI services whose main function is information interaction generally attract fault-based liability, whereas AI products, exemplified by embodied intelligence, that endanger personal and property safety due to product defects may attract strict liability under product liability rules. This distinction is based on the attributes of the product, the nature of the danger and the interests infringed; it does not mean that strict liability applies whenever AI is autonomous or causes damage. Read together with Articles 3 and 9 of the Opinion, in a specific case a court should first examine whether the conditions for applying a special rule are met, and then determine the corresponding mode of attribution.

At the same time, for parties to which fault-based liability applies, the Opinion requires consideration of factors such as the application scenario, the scope of risk, the capacity to foresee and control risk, and the technical feasibility of preventive measures. For example, where a business operator actively selects a celebrity’s image for commercial promotion, authorization and the truthfulness of the publicity should be examined; for a party providing general-purpose technology, the duty of care should be determined in light of its actual participation and its capacity to address risk. Whether the actor has breached a duty and whether it should be liable for all of the damage remain questions to be examined separately.

II. Generative AI: Protection of Rights and Interests and Liability Boundaries in Content Generation

Generative AI affects relevant rights and interests mainly through information processing and content generation. Once agents further acquire the ability to autonomously search, crawl, organize and publish, the risk of infringement may run through the obtaining of material, the processing of content and the dissemination process, rather than existing only in the final output. Articles 4 to 6, 10 and 12 of the Opinion address personality rights, personal information, consumer rights and intellectual property respectively; where the same content involves multiple rights and interests, the conditions for each should be examined separately. A brief analysis follows:

1. Personality rights and interests: identifiability and the scope of authorization

First, AI face-swapping and the use of virtual images. On 5 April 2026, the studio of Yi Yangqianxi stated that AI-generated drama series using his likeness, among others, had been circulated on certain online platforms without authorization, and that he had not acted in the relevant series nor authorized any third party to create AI composites. The key issue in such disputes is that, although the actor did not actually appear, his identifiable image may still be used, through AI synthesis, in the creation and dissemination of dramas. Article 4 of the Opinion takes as its focus of review whether the generated image can identify a specific natural person and whether consent was obtained for its use and disclosure, while preserving the exception that the law may provide otherwise.

In other words, differences between a synthesized character and the person’s own image do not automatically exclude protection of portrait rights. The assessment of identifiability should take into account facial features, names, labels and the context of presentation, and examine whether the general public or a specific group of people can associate the image with the person; it cannot rest on a comparison of facial similarity alone. It may therefore be considered that, in the above case, the unauthorized use and dissemination of an AI-generated portrait of another person constitutes infringement. In addition, even if the synthesized character is labelled as ‘AI-generated’, this only explains how the content was produced and does not prove that the person consented to the use of his image in the relevant storyline; nor can entertainment purposes or non-profit use alone constitute a lawful basis for use.

Second, voice cloning and synthesis likewise require review of the scope of authorization. Article 4 of the Opinion addresses the situation where, without consent, a natural person’s voice is used as training corpus to generate recognizable synthetic voice, and its rights basis is linked to the Civil Code provisions on name rights, portrait rights, voice rights and reputation rights.

In the reference case included in the People’s Courts Case Database, ‘Yin Mouzhen v. Beijing Certain Intelligent Technology Co. et al. (dispute over personality rights)’, a culture company provided Yin Mouzhen’s recording data to a software company for AI processing without his consent, and the latter produced a text-to-speech product that retained the identifiable features of his voice and sold it. The court held that the two companies had used AI technology to process and exploit his voice without lawful authorization, infringing his voice rights and interests. The case indicates that rights relating to sound recordings and a natural person’s voice rights and interests are different objects of protection: obtaining permission to use a sound recording does not automatically include consent to use the voice in it for model training and subsequent synthesis; the authorizing party, the purpose and the scope must still be examined.

Beyond portrait and voice rights and interests, AI face-swapping and voice imitation may also involve reputation rights. Article 4 of the Opinion provides that manipulating a virtual image or voice capable of identifying a specific natural person to commit improper acts or make untrue statements, thereby lowering the social evaluation of that person or another person, shall be determined in accordance with law to constitute an infringement of reputation rights. Accordingly, unauthorized use and damage to reputation should be examined separately.

Third, in special application scenarios such as ‘AI resurrecting the deceased’, the dispute often concerns the use of the deceased’s image to express statements that he or she never made. In June 2026, it was reported that an account, using a photograph of Zhang Xuefeng as its background together with an AI-synthesized voice, produced a video predicting the college entrance examination essay topic, presenting the content as if narrated by him. In such a case, it should be determined whether the newly generated statements are attributed to the deceased, and the basis for their use and the impact on the deceased’s personality interests should be further examined. Article 4 of the Opinion provides that where a virtual digital image of a deceased person is created or used without authorization, causing the deceased’s name, portrait or reputation to be infringed, the close relatives of the deceased may request the actor to bear civil liability under Article 994 of the Civil Code. With respect to such videos, whether the use of the name and portrait is lawful and whether the fabricated statements lower the deceased’s social evaluation should be examined separately; damage to reputation cannot be found merely because the statements were not made by the person himself or herself. Where infringement is established, close relatives may request cessation of the infringement in accordance with law and, depending on the specific consequences, seek elimination of adverse effects or an apology. Even in the name of remembrance or commemoration, therefore, review of the specific content and manner of use cannot be dispensed with.

Fourth, where the source of material and data processing involve personal information, Article 6 of the Opinion permits, within a reasonable scope, the use for model training of personal information that an individual has made public himself or herself or that has otherwise been lawfully made public and to which the individual has not explicitly objected. For example, whether information published by an individual on a public website can be used for training should still be judged in light of the purpose of training, the sensitivity of the information and the scope of use reasonably foreseeable at the time of publication; it cannot be used arbitrarily merely because it has been made public. Where there is a significant impact on the individual’s rights and interests, the individual’s consent should be obtained in accordance with law. For example, the fact that a user uploads a document to a large model merely to revise its wording does not amount to making the information public. If a vendor uses it for training, it should be examined whether there is valid consent covering that use or another lawful basis for processing; Article 6 cannot be applied merely because the user voluntarily uploaded the file. Neither do the above personal information processing rules replace the review of authorization concerning personality rights and interests such as portrait and voice rights.

2. Intellectual property, commercial exploitation and service provider liability

First, as to intellectual property protection, Article 12 of the Opinion requires comprehensive consideration of the type of service, industry characteristics, the source of training data, the degree of participation of each party, the necessary measures taken and the profits obtained, and the liability of the relevant parties to be determined accordingly. The article does not require developers, providers and users to be held jointly liable simply because the generated content is infringing; rather, it emphasizes determining liability based on each party’s own conduct and degree of participation.

As for users, Article 12 makes liability conditional on the user’s knowledge or constructive knowledge of the existence of the prior work, substantial similarity between the generated content and that work, and the absence of a reasonable defence. For example, where a user asks AI to imitate Yu Hua’s writing style and the generated content is similar only in general features such as linguistic rhythm and narrative approach, this is usually not enough to establish infringement; but if it reproduces the original expressive text or the specific plot arrangement of one of his novels, it may still constitute substantial similarity even if character names are changed and some sentences rewritten. The conduct of users who generate content with the help of AI should therefore be analysed by category, distinguishing use of protected expression from non-expressive use: infringement cannot be found merely because the style is similar, nor can the use of AI to create be a ground for exemption from liability.

As for developers raising a non-infringement defence, the court should order them to provide corresponding technical materials in support. For example, suppose a developer trains a model on a certain author’s novels and the model repeatedly outputs content substantially similar to specific passages or plots of the original work; if the developer claims that it has only learned general linguistic patterns, it should provide the source of the training data, records of the training process, the model’s operating mode and the scientific theoretical basis in support of that defence. In other words, the SPC takes the view that such materials assist in examining the basis for the use of the training data and the generation mechanism of the model, and that this data can only be provided by, and is in the custody of, the developer, which should therefore provide the above materials when raising a non-infringement defence.

Second, where generated content is used in commercial marketing, infringement of personality rights and interests and consumer fraud should be distinguished. Article 10 of the Opinion links punitive damages for impersonating celebrities in livestream selling to ‘constituting fraud’, which means that the unauthorized use of a celebrity’s image does not automatically make a consumer’s claim for punitive damages succeed; it is also necessary to examine elements such as whether the operator deliberately fabricated a recommendation relationship and whether the consumer, as a result, formed a mistaken understanding and made a transactional decision. For example, whether the goods are of acceptable quality cannot by itself determine whether fabricating a celebrity recommendation relationship constitutes fraud; the overall promotional conduct must also be considered, that is, if it could lead consumers to believe that the celebrity had authorized or endorsed the goods, fraud is established and the consumer is entitled to claim punitive damages.

Moreover, in practice most operators, especially e-commerce platforms, commonly engage in ‘algorithmic price discrimination against regular customers’. The Opinion expressly provides that an operator’s use of algorithms to apply unreasonable differential treatment in transaction terms such as price may constitute infringement; that is, whether differential treatment is established and whether the reasons for applying it are legitimate and sufficient are important factors in determining whether the operator has infringed.

Third, as to responsible parties and the duty of care of service providers, Article 7 of the Opinion reflects the ‘safe harbour’ approach, that is, delineating the boundary of liability through notice by the right holder and the adoption of necessary measures by the service provider. The article distinguishes two situations: automatic generation and malicious inducement by users. Where automatically generated content infringes personality rights and interests and, after notice, the service provider fails to take necessary measures promptly, it shall be liable for the resulting damage in accordance with law; where a user maliciously induces the generation of infringing content, the user shall be liable in accordance with law, and if the service provider, after notice, fails to deal with it promptly, it shall, under Article 1195 of the Civil Code, be jointly and severally liable with the user for the enlarged part of the damage.

Furthermore, compared with the ‘notice-and-takedown’ measures applied to existing online content, the generative conduct of a generative AI service provider should also extend to stopping generation and blocking the relevant instructions. For example, where a particular instruction can repeatedly generate infringing content and the service provider merely deletes the existing results without controlling the relevant instruction, thereby leaving the risk of subsequent generation, it may be considered that the provider has not fulfilled its disposal obligation in accordance with law. Whether the necessary duty of care has been fulfilled should therefore be judged in light of the content of the notice, the type of service and technical feasibility. The ‘safe harbour’ principle does not mean exemption from liability before notice is received: where the service provider itself participates in the infringement, or knows or should know of the user’s infringement and fails to take necessary measures, liability should still be determined under the relevant provisions.

III. Physical AI: Determination of Product Defects and Allocation of Responsibility Among Multiple Parties

First, as to the identification of AI products, Article 9 of the Opinion requires that, based on the definition of ‘product’ in the Product Quality Law, the corresponding rules be applied to AI products carried in physical form. Accordingly, where AI is embedded in a robot or a vehicle, it should be further examined whether the product is defective and whether the defect is causally related to the damage; where the damage results from improper operation, maintenance or on-site management, the liability of the corresponding actor must still be assessed.

For example, AI products usually have specific applicable scenarios and capability boundaries; a function that fails to operate may be due to use beyond the applicable conditions, or may be a malfunction of a product that should operate normally. The determination of a defect therefore requires a further distinction: whether the functional limitation is a reasonable restriction that has been adequately explained, or constitutes an unreasonable risk to product safety. Article 9 of the Opinion requires consideration of factors such as the nature and use of the product, its autonomous learning capability, upgrades and updates, and the degree of user control over the system, and focuses on whether the producer or seller has provided true explanations and clear warnings regarding the applicable scenarios, inherent limitations and foreseeable risks, thereby providing a basis for judgment.

Two reference cases in the People’s Courts Case Database illustrate the above distinction. In Shen Moumin v. Taizhou Hai Mou Automobile Sales Service Co., Ltd., the accident involved a two-wheeled electric vehicle approaching from the side; the user manual had explained the relevant recognition limitation, and the driver had also continuously pressed the accelerator by mistake. Taking into account the system’s scope of application, the priority mechanism for driving instructions and the cause of the accident, the court dismissed the claim. Shu Mougui v. Huaihua Chun Mou Automobile Sales Service Co., Ltd., Da Mou Sales Co., Ltd. et al. was different: the court commissioned an appraisal institution to conduct road tests of the functions stated in the manual and found that the automatic braking and warning functions failed, with actual performance significantly different from the manual, and on that basis supported the claims for return of the vehicle and refund of the price.

The difference between the two cases lies in this: in the former, the accident situation involved a recognition limitation that the system had already explained, and, taking into account the driver’s operation and the cause of the accident, the court did not support the product liability claim; in the latter, the appraisal found that the system still failed under the functional conditions stated in the manual, so there was a basis for finding a product defect. The focus of the assessment is accordingly whether the product should have functioned under the relevant conditions, whether its actual performance conforms to its functional description, and whether the limitation itself constitutes an unreasonable risk. Even if the operator has disclosed a certain limitation, this does not shift an unreasonable safety risk inherent in the product to the user.

Second, in the scenario of traffic accidents caused by intelligent driving, tort liability needs to be further linked to road traffic accident liability and product liability. Article 11 of the Opinion continues to be based on the Civil Code and the Road Traffic Safety Law: where damage is caused by a vehicle defect, the producer or seller may be requested to bear product liability in accordance with law; where an assisted-driving vehicle causes the same damage through a combination of driver fault and vehicle defect, the respective liabilities may be determined under the Civil Code. In other words, activating assisted driving does not automatically exempt the driver from the duty of care, nor does the occurrence of an accident automatically prove that the system is defective. Takeover prompts, the driver’s response, the operating state of the system and the scope of application of the function may all affect the determination of liability.

Third, where multiple parties participate in the operation of the system, the final product manufacturer, the deployment operator and upstream technology providers must also be distinguished. Some scholars consider that it is significant for the final product manufacturer to bear product defect liability, and that deployment and operation entities directly facing end users are also better placed to prevent damage through permission settings, operational restrictions and risk warnings; upstream developers that do not directly face end users, by contrast, should not be subject to the same liability merely because they supply technology. For example, injury caused by a robot may involve the safety design of the complete machine, or the operator’s site setup and operational management. In asserting the liability of each party, their actual conduct and role should be ascertained; joint and several liability cannot be found simply because they participate together in the technology chain.

In addition, liability concerning vehicle advertising should be examined separately. Article 11 of the Opinion provides for false or misleading publicity regarding automation levels, degree of intelligence, performance and use. Exaggerating assisted driving capabilities may affect consumers’ purchasing decisions and may also cause over-reliance in use; but in asserting the corresponding liability, transaction misrepresentation and accident-caused loss should still be distinguished, and the impact of the publicity on the purchase decision, driving behaviour and damage consequences examined separately. The operator cannot be held liable for the entire accident simply because the publicity was inaccurate.

IV. Practical Outlook

In summary, from content generation to physical operation, the Opinion requires liability to be determined in specific scenarios within the existing legal framework. The former focuses on reviewing authorization, content and the conduct of each participant, while the latter requires ascertaining the relationship between product defects and operational and management factors. Read together with Articles 11 and 12 and Articles 17 to 18 on evidence, training sources, generation records and operational data may become important bases for determining liability, but the relevant evidentiary requirements do not mean that the burden of proof is reversed in all AI cases.

In concrete judicial application, many discretionary standards and issues of legal application remain to be further clarified and refined:

• Determining the boundary of ‘fair use’ for training sets: whether using copyrighted works without permission to train large models constitutes ‘fair use’, and the standard of discretion between protecting original copyright and encouraging technological innovation, remain unclear;
• The threshold for finding ‘originality’ in generated output: the degree of original expression that a user’s prompt must reach for the resulting text, image or video to obtain copyright protection remains subject to considerable disagreement in practice, and the Opinion does not address it for the time being;
• Burden of proof and data retrieval in complex technical scenarios: faced with algorithmic black boxes and highly integrated smart devices, ordinary parties are at a natural disadvantage in proving product defects or technical fault. How to reasonably allocate the burden of proof in litigation, and how to lawfully obtain and preserve underlying operational data and system logs, still await further refinement through subsequent judicial interpretations and typical cases.

References

[1] Supreme People’s Court, Opinion on the Adjudication of AI-Related Disputes in Accordance with Law, Fa Fa [2026] No. 10, issued on 7 September 2026. Official SPC website: https://www.court.gov.cn/zixun/xiangqing/511101.html.

[2] Zhou Jiahai, Si Yanli, Qin Yuanming, Jia Yuhui and Zhang Yin, ‘Understanding and Application of the Opinion of the Supreme People’s Court on the Adjudication of AI-Related Disputes in Accordance with Law’, article on the Legal Application WeChat public account, which states at the beginning that it is intended for publication in the ‘New Laws and New Interpretations’ column of Legal Application, Issue 11 of 2026. Version published by the Intellectual Property Court of the Supreme People’s Court: https://ipc.court.gov.cn/zh-cn/news/view-6039.html.

[3] Supreme People’s Court, ‘Relevant Officials of the Supreme People’s Court Answer Questions from Journalists on the Opinion on the Adjudication of AI-Related Disputes in Accordance with Law’, 7 September 2026. Official SPC website: https://www.court.gov.cn/zixun/xiangqing/511111.html.

[4] Pan Jun, ‘Grandall Viewpoint | Setting Judicial Boundaries for the AI Era - A Systematic Interpretation and Practical Guide to the Supreme People’s Court’s Opinion on the Adjudication of AI-Related Disputes in Accordance with Law’, article on the Grandall Law Firm WeChat public account.

[5] Reference case in the People’s Courts Case Database: ‘Yin Mouzhen v. Beijing Certain Intelligent Technology Co. et al. (dispute over personality rights)’, database No. 2025-07-2-474-001. Case commentary published by the court: https://sdcourt.gov.cn/wfzy/442541/442474/37447542/index.html.

[6] Reference case in the People’s Courts Case Database: ‘Shen Moumin v. Taizhou Hai Mou Automobile Sales Service Co., Ltd. (product liability dispute)’, database No. 2025-07-2-373-007. The facts of the case in this article are taken from Case 5 in the Beijing Lawyers Association’s Guidelines for Lawyers Handling Product Quality Liability Disputes: https://www.beijinglawyers.org.cn/file/202512/202512_1765334139005.pdf.

[7] Reference case in the People’s Courts Case Database: ‘Shu Mougui v. Huaihua Chun Mou Automobile Sales Service Co., Ltd., Da Mou Sales Co., Ltd. et al. (product liability dispute)’, database No. 2025-07-2-373-006. The facts of the case in this article are taken from the case published in the Business Information of the Civil and Commercial Law Professional Committee (August 2025 issue): https://www.xclslawyer.cn/uploads/20250801/99e7b16f57cbdc95c7c9bd9259e29c75.pdf.

[8] ‘Several AI Short-Drama Characters Look Alike Yi Yangqianxi, Studio Statement: Infringing Parties Will Be Held Accountable in Accordance with Law’, The Paper (Pengpai Hao - Media), 5 April 2026. https://www.thepaper.cn/newsDetail_forward_32905488.

[9] ‘AI Resurrects Zhang Xuefeng to Predict College Entrance Examination Essay Topics, Parents of Candidates Call It Absurd! Ministry of Education Issues Examination Alert: Candidates and Parents Should Not Superstitiously Rely on Topic Predictions, So as Not to Disrupt the Pace of Preparation’, Guangming Online Pengpai Hao, 4 June 2026. https://m.thepaper.cn/newsDetail_forward_33305703.

[10] Civil Code of the People’s Republic of China, Articles 994, 1165, 1172, 1195, 1197 and 1202. Policy and Regulation Database of the State Taxation Administration: https://fgk.chinatax.gov.cn/zcfgk/c100009/c5212250/content.html.

[11] Personal Information Protection Law of the People’s Republic of China, Articles 13, 14, 27 and 69. Text published by the State Administration for Market Regulation: https://www.samr.gov.cn/wljys/gzzd/art/2023/art_3ef1e889c1e644d4b65b5f5c7f432386.html.

[12] Guiding Case No. 81 of the Supreme People’s Court: Zhang Xiaoyan v. Lei Xianhe, Zhao Qi and Shandong Aishuren Audio-Visual Books Co., Ltd. (copyright infringement dispute). Cited to illustrate the object of comparison for substantial similarity of literary works; it is not an AI-related case. https://www.court.gov.cn/shenpan/xiangqing/37642.html.

[13] Wang Liming, ‘On the Determination of Tort Liability in the Application of AI Agents’, Tsinghua Law Review, Issue 5 of 2026, reprinted on the WeChat public account of China Civil and Commercial Law Network. This article mainly refers to Part I, ‘The Particularity of Legal Interests Infringed by Agents and the Response’, Part II, ‘The Plurality of Principles for Attributing Agent Infringement Liability and Their Determination’, and Part III, ‘The Diversity of Responsible Parties for Agent Infringement Liability and the Determination of Their Liability’.

Appendix: Full Text

Fa Fa [2026] No. 10 - Opinion of the Supreme People’s Court on the Adjudication of AI-Related Disputes in Accordance with Law

To comprehensively implement the spirit of the 20th National Congress of the Communist Party of China and all its plenary sessions, to serve and safeguard the accelerated development of new quality productive forces through strict and impartial justice, and to guide people’s courts at all levels in adjudicating AI-related disputes in accordance with law, the following Opinion is hereby issued in accordance with the Civil Code of the People’s Republic of China, the Cybersecurity Law, the Data Security Law, the Copyright Law, the Anti-Unfair Competition Law, the Law on the Protection of Consumer Rights and Interests, the Personal Information Protection Law, the Civil Procedure Law and other relevant laws, and in light of the actual trial work of people’s courts.

I. General Requirements

(1) Guiding ideology

Guided by Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era, thoroughly implementing Xi Jinping Thought on the Rule of Law, earnestly implementing General Secretary Xi Jinping’s important thought on building a strong cyber nation, and deeply recognizing that AI is an important driving force of the new round of scientific and technological revolution and industrial transformation, we shall attach equal importance to development and security and combine the promotion of innovation with law-based governance, adjudicate AI-related disputes fairly in accordance with law, promote the safety, reliability, controllability and fairness of AI technology applications, and provide strong judicial services and safeguards for comprehensively advancing the building of a strong nation and the great cause of national rejuvenation through Chinese-style modernization.

(2) Basic principles

Adhere to a people-centred approach. Give play to the guiding role of judicial adjudication, actively guide AI to enhance people’s wellbeing, and effectively strengthen judicial protection of people’s livelihood rights and interests in the digital era. Actively and prudently explore and improve the construction of judicial adjudication rules and mechanisms in AI fields such as algorithmic ethics and algorithmic rules, make full and good use of existing legal provisions, properly respond to rule conflicts, social risks and ethical challenges brought by the development of AI technology, promote the improvement of corporate social responsibility systems for AI enterprises, focus on preventing and eliminating algorithmic discrimination, protect individuals’ right to know and right to choose in automated decision-making in accordance with law, prevent and regulate the use of AI technology to infringe the lawful rights and interests of the people, and promote AI that serves the people and works for good.

Support innovation and development. Respect the laws of scientific and technological innovation and the development practice of the AI industry, and support scientific and technological innovation and industrial innovation through judicial adjudication that conforms to the spirit of laws and policies and the laws of technological development. Explore judicial protection paths for new types of rights and interests, and continuously increase protection of AI innovation achievements, innovation entities, innovation conduct and the innovation environment. Balance the protection of rights and interests with the development of the technology industry in an inclusive and prudent manner, and create a favourable environment that encourages exploration and tolerates failure. Regulate monopoly or unfair competition carried out using AI technology in accordance with law, protect in accordance with law the equality of rights, opportunities and rules of relevant entities in developing and using AI technology, and strengthen equal protection of the rights and interests of business entities such as small and medium-sized enterprises. By adjudicating AI-related disputes fairly and efficiently in accordance with law, actively guide and promote the deep integration of the internet, big data, AI and the real economy, help cultivate and expand the AI industry, and maximize the huge potential of AI to empower high-quality economic and social development.

Firmly safeguard the bottom line of security. Based on the specific damage that AI may cause to the lawful rights and interests of civil subjects in different application scenarios and the nature and magnitude of the risk, accurately determine legal liability in accordance with law. Distinguish the differences among general and specialized, open-source and closed-source large models in terms of technical principles, risk spillover and control capacity, and reasonably allocate the legal liability of developers, providers, users and other entities. Comprehensively and accurately implement the criminal policy of balancing leniency and severity: resolutely punish in accordance with law criminal acts such as endangering national security, infringing citizens’ rights and interests, and disrupting social order committed by deliberately using AI; and handle innovative conduct in AI research, development and application prudently in accordance with law, abiding by the principles of legality of crime and punishment and suiting punishment to crime and responsibility.

II. Properly Adjudicate AI-Related Tort Cases in Accordance with Law and Effectively Safeguard the Lawful Rights and Interests of Civil Subjects

(3) Accurately grasp the principles for attributing AI-related tort liability. Accurately apply the Civil Code, the Personal Information Protection Law, the Product Quality Law, the Road Traffic Safety Law and other relevant legal provisions, and determine in accordance with law the legal liability arising from the use of AI to infringe civil rights and interests. Where the law does not clearly provide for no-fault liability or liability based on presumption of fault, whether the actor bears tort liability shall be determined under the fault liability principle provided in Article 1165(1) of the Civil Code. In judging whether an actor is at fault and the degree of fault, comprehensive consideration shall be given to factors such as the specific scenario of the AI application, the degree of autonomy, the transparency of technology and information, the potential risk and scope of impact, the measures taken by AI developers, providers and other relevant entities to prevent and reduce AI infringement and the technical feasibility thereof, and the AI user’s capacity to foresee and control the damage that may be caused by the use of AI technology to commit infringement.

(4) Regulate in accordance with law acts such as ‘AI face-swapping and voice imitation’ and ‘AI resurrecting the deceased’ that use generative AI to infringe personality rights and interests. The use of generative AI to process the name, portrait, etc. of a specific natural person or deceased person shall not violate laws or regulations or public order and good morals. Unless otherwise provided by law, where, without the natural person’s consent, AI is used to process the natural person’s name, portrait, etc. to generate a virtual digital image capable of identifying that natural person and to use or disclose it, and the natural person claims that the actor has infringed his or her name rights, portrait rights or other personality rights and interests, the people’s court shall support the claim in accordance with law. Unless otherwise provided by law, where, without the natural person’s consent, the natural person’s voice is used as training corpus to imitate the timbre, intonation and pronunciation style of that natural person to generate synthetic voice capable of identifying that natural person, and the natural person claims that the actor has infringed his or her voice rights and interests, the people’s court shall support the claim in accordance with law. Where a generated or synthesized virtual digital image or voice capable of identifying a specific natural person is manipulated to commit improper acts or make untrue statements, thereby lowering the social evaluation of that natural person or another person, the people’s court shall determine in accordance with law that infringement of the right of reputation is constituted. Where AI technology is used to create or use the virtual digital image of a deceased person without authorization, causing the deceased person’s name, portrait, reputation, etc. to be infringed, and the close relatives of the deceased request the actor to bear civil liability under Article 994 of the Civil Code, the people’s court shall support the request in accordance with law.

(5) Regulate in accordance with law acts such as ‘online doxxing’ and ‘human flesh search’ carried out using AI that infringe natural persons’ right to privacy. Where, for the purpose of prying into privacy, AI is used to track and analyse public information such as a specific natural person’s telephone number, online account and social media, to obtain private information, or the obtained private information is leaked or disclosed, or the obtained information is used to disturb the peace of private life, this shall be determined to constitute infringement of the right to privacy. Where AI is used to photograph, peep at, eavesdrop on, etc. a specific natural person’s private space or private activities, this shall be determined to constitute infringement of the right to privacy, unless otherwise provided by law or the natural person has explicitly consented.

(6) Determine in accordance with law civil liability for infringement of personal information rights and interests in the AI training process. Where, for AI model training, personal information that the individual has made public himself or herself or that has otherwise been lawfully made public is processed within a reasonable scope and the individual has not explicitly objected, this is generally not determined to be an act infringing personal information rights and interests. Where there is a significant impact on the individual’s rights and interests, the individual’s consent shall be obtained in accordance with legal provisions. In determining a reasonable scope, comprehensive consideration shall be given to factors such as the necessity and appropriateness of the purpose of processing personal information and the model functions, the type and sensitivity of the personal information involved and its potential impact on the individual’s rights and interests, and the scenario in which the individual made the information public and the reasonably foreseeable scope of use.

(7) Prudently determine the tort liability of generative AI service providers. Where content automatically generated by generative AI infringes others’ reputation rights, privacy rights or other personality rights and interests, and, after notice by the right holder, the generative AI service provider fails to promptly take necessary measures such as stopping the generation of the infringing content, it shall bear tort liability for the resulting damage in accordance with law. The notice shall include preliminary evidence of the infringement and the true identity information of the right holder. Where a network user maliciously induces generative AI to generate infringing content by inputting infringing prompts and thereby causes damage to others, that network user shall bear tort liability in accordance with law; where, after notice by the right holder, the generative AI service provider fails to promptly take necessary measures such as stopping the generation of the infringing content and blocking the relevant generation instructions, and the right holder requests the network user and the generative AI service provider to bear civil liability under Article 1195 of the Civil Code, the people’s court shall support the request in accordance with law.

(8) Apply the personality rights infringement injunction system in accordance with law. Where a natural person, legal person or unincorporated organization has evidence proving that an actor is using or is about to use AI to carry out unlawful acts infringing its personality rights and interests, and that failure to stop them promptly will cause irreparable damage to its lawful rights and interests, and applies to the people’s court for measures ordering the actor to stop the relevant conduct, or ordering the relevant network service provider or generative AI service provider to stop providing the relevant services, the people’s court may issue a personality rights infringement injunction in accordance with law. When issuing such an injunction, the people’s court shall, taking into account the type of personality right infringed, the manner of the unlawful act and the scope and degree of possible damage, adopt corresponding injunctive measures and shall not exceed the necessary limits.

(9) Accurately determine AI product liability. People’s courts shall, based on the definition of ‘product’ in the Product Quality Law, accurately identify AI products carried in physical form in accordance with law and apply the corresponding legal rules. Where an AI product is defective and causes damage, the producer and seller shall bear product liability in accordance with law. In determining whether an AI product presents an unreasonable danger endangering personal and property safety, comprehensive consideration shall be given to factors such as the nature and use of the AI product, its autonomous learning capability, upgrades and updates, the degree of user control over the system, and whether it conforms to relevant national and industry standards, and particular attention shall be paid to whether the producer and seller have provided true explanations and clear warnings regarding the product’s applicable scenarios, inherent limitations and foreseeable risks.

(10) Regulate in accordance with law acts infringing consumers’ lawful rights and interests, such as ‘algorithmic price discrimination’ and ‘impersonating celebrities in livestream selling’. Where, for the same goods or services, an operator uses algorithms to apply unreasonable differential treatment in transaction terms such as price and thereby infringes others’ lawful rights and interests and causes damage, the people’s court shall determine in accordance with law that it bears corresponding tort liability. In determining whether unreasonable differential treatment is constituted, comprehensive consideration shall be given to whether the differential treatment substantially restricts or harms consumers’ right to know, right to independent choice, right to fair trade, etc., whether it forms personalised transaction terms based on information such as consumers’ consumption preferences, willingness and ability to pay and browsing records, and whether it violates the principle of good faith and business ethics, so as to assess whether the reasons for the differential treatment are legitimate, sufficient and non-discriminatory. Where an operator, in providing goods or services, uses AI to ‘impersonate celebrities in livestream selling’ and this constitutes fraud, and the consumer claims punitive damages under Article 55 of the Law on the Protection of Consumer Rights and Interests, the people’s court shall support the claim in accordance with law.

(11) Determine in accordance with law liability for damages in traffic accidents involving autonomous vehicles and vehicles with assisted driving functions. Where an autonomous vehicle or a vehicle with assisted driving functions is driven on the road and causes damage in a traffic accident, liability for damages shall be borne in accordance with the relevant provisions of the Civil Code and the Road Traffic Safety Law. Where damage is caused by a traffic accident due to a vehicle product defect and the party requests the producer or seller to bear liability for damages under Chapter IV of Part VII of the Civil Code, the people’s court shall support the request. Where a vehicle with assisted driving functions causes the same damage through a combination of a vehicle defect and the driver’s faulty conduct, and the party, under Article 1172 of the Civil Code and other provisions, simultaneously requests the driver and the vehicle producer or seller to bear liability for damages, the people’s court shall support the request in accordance with law. Where a vehicle producer or seller makes false or misleading publicity about the automation level, degree of intelligence, performance or use of an autonomous vehicle or a vehicle with assisted driving functions, harming consumers’ lawful rights and interests, and the consumer requests the vehicle producer or seller to bear civil liability under the Civil Code, the Law on the Protection of Consumer Rights and Interests and other legal provisions, the people’s court shall support the request in accordance with law. To ascertain the cause of a road traffic accident, the people’s court may require the vehicle producer, seller or operator and other data controllers to provide, within the necessary scope, true and complete autonomous driving and assisted driving event records and other data needed to ascertain the facts of the case.

III. Properly Adjudicate AI-Related Intellectual Property Disputes in Accordance with Law and Safeguard High-Quality Development and High-Level Security of AI

(12) Regulate in accordance with law acts of infringing intellectual property rights involving AI. Where AI-generated content infringes another’s copyright, the people’s court shall, taking comprehensive account of factors such as the type of AI service, industry characteristics, the source of training data, the degree of participation of each party, the necessary measures taken and the profits obtained, reasonably determine in accordance with law the liability of the AI developer, provider and user. Where an AI developer raises a non-infringement defence, it shall be ordered to provide the source of the training data, records of the training process, the model’s operating mode and the scientific theoretical basis in support. Where a right holder claims that an AI provider has used algorithmic technology to infringe its copyright, it shall provide the relevant evidence. Where an AI user knows or should know of the existence of a prior work and uses AI to generate a work substantially similar to the prior work without a reasonable defence, and the right holder of the prior work requests it to bear infringement liability, the people’s court shall support the request. Where AI is used to commit acts such as counterfeiting, false advertising, or fake traffic and fake orders, infringing others’ rights or constituting unfair competition, the corresponding liability shall be borne in accordance with law.

(13) Reasonably determine in accordance with law the legal liability of parties involved in AI-related open-source software. In adjudicating cases involving open-source software, in determining the infringement liability of open-source software developers and providers and of subsequent developers and providers, comprehensive consideration shall be given to factors such as the type of open-source licence agreement, the specific content of rights restrictions, security and compliance measures and the degree of information disclosure, and appropriate liability exemption shall be granted to open-source software developers and providers in accordance with law. Where an open-source software developer or provider provides, on a free and open-source basis, certain code modules required for AI-related software research and development and publicly explains their functions and security risks, and another person’s use of the code module causes infringement, the people’s court may determine that the open-source software developer or provider does not bear infringement liability.

(14) Regulate in accordance with law patent grant and validity confirmation conduct concerning AI-related inventions and creations. Where an AI-related invention or creation adopts technical means following natural laws, solves a technical problem and achieves a technical effect conforming to natural laws, the people’s court shall determine in accordance with law that it is subject matter protected by the Patent Law, unless it violates laws, social morality, harms the public interest, or the natural person has made no substantial contribution. Where a natural person completes an invention or creation using AI and the natural person has made a creative contribution to the substantive features of the invention or creation, that natural person shall be recognized as the inventor. Where the description of an AI-related patent expresses the technical solution to a degree that a person of ordinary skill in the art can implement the invention, it shall be determined to meet the conditions for authorization of sufficient disclosure.

(15) Prudently determine in accordance with law liability for breach of contract in the performance of AI-related technology contracts. In adjudicating disputes over AI-related technology development, transfer, licensing, consulting and service contracts, people’s courts shall, in accordance with the contract and taking full account of the characteristics of AI technology research and development and whether the technology developer has made reasonable efforts, determine liability for breach of contract in accordance with law.

(16) Regulate in accordance with law AI-related data use conduct. Where an AI developer lawfully obtains data through collection and generation, derivative creation, assignment, licensing, etc. and enjoys corresponding data rights and interests, the people’s court shall protect them. Data and data collections that constitute compilations or meet the elements of other works shall be protected under the Copyright Law. Data and data collections that constitute trade secrets shall be protected under the Anti-Unfair Competition Law. For data and data collections that do not constitute trade secrets, where the alleged infringing conduct violates Article 13 of the Anti-Unfair Competition Law, liability shall be borne in accordance with law. Where an operator uses data, algorithms and other technical means to reach a monopoly agreement or abuse a dominant market position, it shall bear the corresponding liability in accordance with law. Where technical means such as fabricating interfering data, maliciously labelling data or adversarial example attacks are used to harm the operational security of AI, the corresponding liability shall be borne in accordance with law.

IV. Improve Procedural Rules for Adjudicating AI-Related Disputes and Punish in Accordance with Law Illegal and Criminal Acts Such as Obstructing Judicial Order by Using Generative AI

(17) Accurately ascertain the facts of AI-related disputes in accordance with law. Given the highly technical and specialised nature of AI-related disputes, people’s courts shall strengthen litigation guidance and clarification in accordance with law and guide the parties to complete their evidence submission in a timely and comprehensive manner regarding the disputed facts. Evidence that a party cannot collect on its own for objective reasons may be investigated and collected upon application to the people’s court; where necessary, the people’s court may investigate and collect evidence ex officio. Where a party applies for evidence preservation to fix key technology, the people’s court shall promptly examine the application in accordance with law. Where a party controlling documentary evidence, electronic data or other evidence refuses to submit it without justified reason, and the opposing party claims that the content of that evidence is unfavourable to the controlling party, the people’s court may find that claim established. For professional issues involving AI technical principles and operating mechanisms, the people’s court shall, in accordance with law, give full play to the role of people’s assessors, appraisers, expert assistants and technical investigation officers in assisting in ascertaining the facts of the case.

(18) Explore and improve evidence review rules for AI-related disputes. People’s courts shall fully consider the complexity of AI technology, the opacity of system operation and the particularity of evidence collection, and continuously improve evidence review rules adapted to new technological development. Focus shall be placed on reviewing the authenticity and integrity of the processes of generating, collecting, storing and transmitting electronic data. Differentiated evidentiary review priorities shall be clarified for evidence produced by different information technologies such as big data and blockchain. For evidence in the form of big data analysis reports, focus shall be placed on reviewing the source of the underlying data, the cleaning rules and the scientific nature of the analysis methods; for blockchain evidence preservation, focus shall be placed on reviewing the authenticity of data before it is recorded on the chain and the reliability of the technical platform. Where a party uses AI-generated content as evidence of infringement, the people’s court shall determine it by comprehensively considering factors such as the design of the prompt and its influence on the generated result, the degree of similarity between the generated content and the work for which rights are claimed, the consistency of repeated tests, and the model training, algorithm design and generated-content filtering mechanisms.

(19) Regulate in accordance with law acts such as improper evidence collection using AI that obstruct the judicial order. Where a party, taking advantage of the autonomous learning and autonomous decision-making characteristics of AI, obtains false evidence through human intervention or misleading means such as deleting or tampering with labels on generated synthetic content, inputting specific instructions, selectively presenting results or adversarial interference, fabricates the basic facts of a civil case, conducts a false lawsuit and attempts to infringe national interests, the public interest or others’ lawful rights and interests, the people’s court shall reject its claims in accordance with law and impose fines or detention depending on the circumstances; where a crime is constituted, criminal liability shall be pursued in accordance with law. Where a participant in proceedings or another person forges evidence using AI and obstructs the people’s court in hearing a case, it shall be handled under Article 114 of the Civil Procedure Law. Where materials such as litigation documents and case retrieval reports submitted by participants in proceedings are generated using AI, they shall carefully verify the authenticity and accuracy of the relevant laws, judicial interpretations and cases before submitting them to the court, explain the use of AI assistance when submitting them to the court, and bear liability in accordance with law for the authenticity and accuracy of the relevant content.

(20) Punish in accordance with law criminal activities carried out using AI. Where acts such as using AI to commit fraud, insult, defamation, damage commercial reputation or product reputation, infringe citizens’ personal information, illegally obtain computer information system data, or produce, traffic in or disseminate obscene materials constitute a crime, criminal liability shall be pursued in accordance with law. Where an actor, after activating the assisted driving function, uses privately installed accessories to evade monitoring by the assisted driving system, causing a road traffic accident and constituting a crime, criminal liability shall be pursued in accordance with law.

V. Strengthen Trial Guidance for AI-Related Disputes and Continuously Improve Working Mechanisms

(21) Give full play to the role of diversified dispute resolution mechanisms. People’s courts shall actively use diversified dispute resolution mechanisms to properly resolve AI-related disputes, promote the establishment of coordinated mechanisms with competent AI industry authorities, industry-specific and professional mediation organizations, AI experts and scholars, jointly carry out dispute resolution work, and ensure prior mediation is substantive. Based on the function of judicial adjudication, promote the law-based level of prevention and resolution of AI-related disputes, clarify rights and obligations and delineate boundaries of conduct through open trials, explaining the law through cases and other means, give play to the demonstrative and guiding role of judicial adjudication, and promote the source prevention and timely resolution of AI-related disputes.

(22) Strengthen trial supervision and guidance. Actively regulate the application of upgraded jurisdiction, strengthen the demonstrative and guiding role, and ensure substantive trial supervision and guidance. AI-related disputes involving major interests, difficult, complex and new types, significance for establishing rules, or the need to unify the standard for applying the law shall be heard at a higher level by the people’s court at the next higher level. Make full use of the People’s Courts Case Database, strengthen case guidance work, and promote the unification of legal application and the improvement of AI governance rules.

(23) Strengthen departmental collaboration and coordination. Through various means such as issuing judicial suggestions and establishing cooperation mechanisms, give full play to judicial functions and serve national and social governance. Strengthen coordination and collaboration with cyberspace, public security, procuratorial and market regulation departments, and establish and implement long-term comprehensive governance mechanisms. Guide relevant enterprises, platforms and industry associations to regulate and govern themselves, and promote the formation of a whole-chain protection synergy of AI industry self-discipline, administrative supervision and judicial safeguards.

(24) Strengthen international judicial exchange and cooperation. Properly adjudicate AI-related and cross-border data disputes, and, on the basis of the principles of mutual respect, equality and mutual benefit, advance exchange and cooperation in foreign-related adjudication in the AI field, strengthen the alignment and mutual learning of rules, promote the building of a fairer, more reasonable, inclusive and open AI governance system, and promote and protect the diversity of the AI ecosystem and human civilisation.

Supreme People’s Court

7 September 2026