Interpretation of "Opinions of the Supreme People's Court on the Legal Handling of Dispute Cases Involving Artificial Intelligence" - The Liability Attribution Principle for Infringement Liability in Cases Involving Artificial Intelligence
Preface
The rapid iteration of artificial intelligence technology has led to an increasing number of disputes involving artificial intelligence. Judicial practice has accumulated some judicial rules for disputes related to artificial intelligence, but they are scattered across various typical cases and archived cases. On September 7, 2026, the Supreme People's Court issued the "Opinions on the Proper Handling of Disputes Involving Artificial Intelligence" (Document No. 10 of 2026) (hereinafter referred to as the "Opinions"), which is the first judicial interpretation in China regarding disputes involving artificial intelligence, and has significant guiding significance for the trial of disputes involving artificial intelligence in China. Based on the practical experience of providing legal services to clients, I will focus on interpreting the provisions of the "Opinions" regarding the attribution principle of tort liability for infringement involving artificial intelligence.
Original text of the "Opinions" Article 3:
"(3) Accurately grasp the attribution principle for tort liability involving artificial intelligence. Accurately apply relevant laws such as the Civil Code, the Personal Information Protection Law, the Product Quality Law, and the Road Traffic Safety Law, and legally determine the legal liability arising from the infringement of civil rights through the use of artificial intelligence. If there is no clear legal provision for the application of no-fault liability or fault presumption liability, the liability of the actor for infringement should be determined in accordance with the fault liability principle stipulated in Article 1165, Paragraph 1 of the Civil Code. When judging whether the actor is at fault and the degree of fault, comprehensive consideration should be given to the specific scenarios of artificial intelligence application, the degree of autonomy, the transparency of technology and information, potential risks and the scope of impact, the measures and technical possibilities taken by the relevant entities such as the developers and providers of artificial intelligence to prevent and reduce infringement by artificial intelligence, and the foreseeability and control ability of the users of artificial intelligence regarding the damage caused by the infringement using artificial intelligence technology."
Summary interpretation:
Article 3 of the "Opinions" is a general provision on the attribution principle, clarifying that different attribution principles should be applied to different types and scenarios of infringement involving artificial intelligence; for disputes involving infringement by artificial intelligence, if there is no other specific provision in the law, the principle of fault liability as stipulated in Article 1165, Paragraph 1 of the Civil Code should be applied; for disputes involving infringement of personal information rights through the use of artificial intelligence technology, the principle of fault presumption liability should be applied.
Artificial intelligence has a wide range of applications, and the provisions of this article reflect the judicial concept of classifying and adjudicating disputes involving artificial intelligence; artificial intelligence technology is still in a stage of rapid development and iteration. If the principle of no-fault liability or fault presumption liability is widely applied to disputes involving infringement by artificial intelligence, it will not only be detrimental to the guiding role of judicial adjudication for the artificial intelligence industry, but also does not conform to the current development trend of the artificial intelligence industry.
Article 9 of the "Opinions" also clearly states that "The People's Court shall, based on the definition of 'product' as stipulated in the Product Quality Law, lawfully determine the artificial intelligence products that are in the form of physical objects and apply the corresponding legal rules." Artificial intelligence products that apply product liability rules should have a physical carrier, have been processed and manufactured, and the producer should have direct control over the production process of the artificial intelligence products, and have a certain reasonable expectation regarding the quality of the products.
Article 1202 of the Civil Code stipulates: "If a product is defective and causes harm to others, the producer shall bear tort liability." Article 46 of the Product Quality Law stipulates: "Defects refer to the unreasonable risks that products pose to the safety of human life and other people's property; if there are national or industry standards for ensuring human health and personal or property safety, defects refer to those that do not meet the standards." If damage is caused to others due to product defects, the producer shall bear liability without fault.
Artificial intelligence products in the form of physical objects can enter the regulatory framework of the Product Quality Law. However, in cases of infringement caused by artificial intelligence products, the following elements should be considered separately: whether the artificial intelligence product has defects, whether the use of the artificial intelligence product has caused damage, and whether there is a causal relationship between the defect of the artificial intelligence product and the damage. Finally, the liability subject and the scope of compensation should be determined.
Fault presumption liability situation
Article 69 of the Personal Information Protection Law stipulates that if the processing of personal information infringes upon the rights and interests of personal information and causes damage, if the personal information processor cannot prove that it was without fault, it shall bear liability for damages and other tort liabilities.
In the fourth case of the typical cases involving artificial intelligence in the Beijing Internet Court - Liao v. a certain technology and culture company for network infringement liability dispute, 1 a certain technology and culture company, without Liao's authorization or consent, produced a series of videos featuring Liao and made them into a face-swapping template, which was then provided to users for use and to make profits. This process involved the collection, use, and analysis of Liao's personal information, and it was a processing of Liao's personal information. The unauthorized "AI face-swapping" processing of videos containing others' portraits constitutes an infringement of the rights and interests of others' personal information.
Since the personal information processor has the methods for processing personal information, it is difficult for the personal information right holder to obtain them. For cases of infringing upon personal information rights using artificial intelligence technology, it is required that the personal information processor provide evidence to prove that it was without fault, and the principle of fault presumption is applied.
02 Basic principles of tort liability attribution - Fault liability
The first part of this article lists two typical situations where there are clear legal provisions for the application of no-fault liability and fault presumption liability. In addition to the provisions of the law, for cases involving infringement by artificial intelligence technology, the principle of fault liability stipulated in Article 1165, Paragraph 1 of the Civil Code should be applied. Only when the actor infringes upon the civil rights of others and causes damage shall the tort liability be borne.
Before the release of the "Opinions", the Beijing Internet Court in the first case of "AI illusion" infringement in the country 2 discussed whether generative artificial intelligence technology constitutes a product. The Beijing Internet Court held that "a product" should have the following characteristics: first, the product should have a specific and particular purpose, and second, it should have a reasonable and feasible quality inspection standard. However, "generative artificial intelligence has diverse application scenarios, often does not have a specific and particular purpose, is still in a continuous changing and updating process, and presents the characteristic of meeting personalized services. If the defect is discussed abstractly without a specific purpose, it may not only lead to ambiguous liability determination but also hinder technological innovation due to overgeneralization. Therefore, generative artificial intelligence should be defined as a service rather than a product; Meanwhile, at present, there are no reasonable and feasible quality inspection standards for generative artificial intelligence. It is difficult to determine whether there are "defects" in it through clear standards and inspection methods, and thus it is difficult to apply product liability. Therefore, generative artificial intelligence lacks the necessary conditions for applying product liability.
03 Determination of Fault
Compared to other types of infringement, it is more difficult to determine the fault of each participating party in cases of artificial intelligence infringement. One of the reasons is the complexity of artificial intelligence technology itself. Moreover, the author believes that the criteria for determining fault will continuously adjust dynamically with the development of artificial intelligence technology and will not remain unchanged forever. Specifically for individual cases, different case circumstances will also affect the determination of the degree of fault of the involved parties.
Currently, Article 3 of the "Opinions" lists six reference factors:
First, the specific scenarios of artificial intelligence application. Generally speaking, the higher the danger of the application scenario and the higher the requirements for the attention obligations of the relevant parties, the higher the attention obligations will be. For example, applying artificial intelligence technology in the field of autonomous driving is more demanding than applying it in the field of simple image generation. If an actor uses artificial intelligence technology to commit an infringement in the field of autonomous driving, it will cause more serious infringement consequences.
Second, the degree of autonomy of artificial intelligence. Usually, if the autonomy of artificial intelligence is higher, the control degree of the user over the artificial intelligence is lower, and thus the attention obligations to the user are lower. Because when the user cannot control it, imposing an excessively high attention obligation on them is also unreasonable.
Third, the transparency of technology and information. Due to the complexity of artificial intelligence technology, the general users of artificial intelligence services do not have as high an understanding of the technical principles and the reliability of the output of the artificial intelligence as the developers and providers. Therefore, the attention obligations to the developers and providers as the developers and providers of artificial intelligence services are relatively higher. For example, the "Deep Synthesis Management Regulations for Internet Information Services" and the "Method for Identifying Artificial Intelligence Generated and Synthesized Content in Cybersecurity Technology" require providers to make explicit and implicit identifications of the content generated by artificial intelligence.
Fourth, potential risks and the scope of influence. Artificial intelligence services can have a larger scope of influence and higher potential risks. Therefore, the attention obligations to the developers, providers, etc. are higher. If the potential influence range is limited to a single individual, the controllability and the resulting damage outcome are relatively predictable. If it is for an unspecified majority, the scope and degree of the harm are difficult to estimate, and the attention obligations to the actor will be higher.
Fifth, the measures and technical possibilities taken by the relevant parties such as artificial intelligence developers and providers to prevent and reduce artificial intelligence infringement. Currently, the upgrade and iteration speed of artificial intelligence models worldwide is very fast. Often, one company's model leads several months later, and another company releases a more advanced version. This year's artificial intelligence hallucination has decreased significantly compared to last year's hallucination. As technology becomes more mature, the attention obligations to the developers and providers will be higher. Conversely, if the technology itself cannot be achieved, the law will not force the parties to take excessive attention obligations. The author believes that currently, artificial intelligence technology is still in a stage of rapid development. At present, from the perspective of encouraging scientific and technological innovation and technological innovation, the policy orientation will not impose higher attention obligations on developers and providers. However, as technology continues to mature and develop, it is necessary to ensure that the technical capabilities and attention obligations are appropriate, which is conducive to protecting individual rights and balancing technological development.
Sixth, the foreseeing ability and control ability of artificial intelligence users regarding the damage that may be caused by the infringement committed by using artificial intelligence technology. If the damage caused is completely beyond the foreseeing ability and control ability of the user, the degree of fault is relatively lower. Because currently, artificial intelligence technology still has parts that cannot be explained, especially for users, when the actor is completely unable to predict the type of damage, it is not necessary to impose an excessively high attention obligation.
Conclusion
Due to the rapid development of current artificial intelligence technology, there is diversity in the infringement behaviors and manifestations related to artificial intelligence. This provision restates the liability principle of the Tort Liability Chapter of the Civil Code, clearly stating that in cases involving infringement by artificial intelligence, the principle of fault liability is the main approach, with exceptions being no-fault liability and fault presumption liability. This is helpful in encouraging technological innovation and the rapid development of artificial intelligence technology. However, the author believes that as the application of artificial intelligence technology becomes more widespread, more diverse infringement behaviors will emerge in the future. With the continuous advancement and maturity of technology, it is also necessary to require the developers and providers of artificial intelligence technology to assume appropriate duty of care.
Footnote:
1. WeChat official account Beijing Internet Court, Beijing Internet Court's typical cases involving artificial intelligence: https://mp.weixin.qq.com/s/E-9ECMK1t8znrXGAMR1Arg;
2. Hangzhou Internet Court (Case No. 2025-Zhong-018143 of Civil Judgment).
