On September 7, 2026, China's Supreme People's Court ("SPC") issued the Opinions of the Supreme People's Court on the Trial of Cases Involving Artificial Intelligence Disputes in Accordance with the Law (the "Opinions"), the first judicial guidelines addressing disputes relating to artificial intelligence ("AI").[1]

The Opinions comprise five sections and 24 articles, providing judicial guidelines on a number of cutting-edge legal issues relating to AI, including the general framework for AI-related tort liability and AI product liability, copyright infringement involving AI-generated content ("AIGC"), protection of personality rights, rules governing the use of publicly available personal information for AI model training, liability of generative AI ("GenAI") service providers, patent protection for AI-assisted inventions, and liability arising from autonomous and assisted driving systems.

1. General Principle for AI-Related Liability (Article 3)

Article 3 of the Opinions provides that liability for AI-related infringement should be determined based on the specific application of the AI system and its corresponding laws and regulations, such as the Civil Code, Personal Information Protection Law, Product Quality Law, and Road Traffic Safety Law. Where the existing legislation expressly provides for no-fault liability or presumed-fault liability, those rules apply; otherwise, the general principle of fault-based liability will apply to AI-related infringements.

Article 3 further provides that, when determining whether the alleged infringer is at fault and the degree of such fault, courts shall consider the following factors. The SPC Explanatory Article explains these factors as follows:

• The scenario in which the AI system is deployed and used — For example, the AI system used to generate text or images presents different risks from the AI system that is used in autonomous driving or medical diagnosis, and therefore, the applicable duty of care should differ accordingly.

• The level of AI autonomy — Generally speaking, the higher the level of autonomy, the greater the duty of care that would be expected from AI developers and service providers relative to AI users.

• Technical and informational transparency — In particular, whether AI developers or AI service providers have clearly warned users about the technical limitations and relevant legal risks relating to the AI system would be important.

• Potential risks and the scope of their impact — The greater the potential risk and the wider the impact that the AI system will bring to the public, the higher the corresponding duty of care that AI developers and AI service providers shall undertake.

• Preventive measures and their technical feasibility — Where current technology does not make it possible for GenAI service providers to completely prevent infringing AIGC outputs, they generally should not be required to bear a results-based review obligation.

• The user's ability to foresee and control the harm — As AI systems may have a degree of autonomy and opacity, harm may exceed the user's foreseeable range or control, which fact should be taken into account when assessing the user's fault.

This liability framework will be particularly important in future AI litigation cases, involving AI developers, AI deployment providers, AI service providers and AI users.

2. AI-Products Liability v. AI Services Liability (Article 9)

The Opinions draw an important distinction between AI-services liability and AI-products liability.

Article 9 of the Opinions provides that, for the purposes of determining product liability, an "AI product" should generally refer to a product embodied in a physical carrier. The SPC Explanatory Article cites "embodied-intelligence robots" and "autonomous vehicles" as examples of such AI-enabled physical products, and clarifies that AI services themselves do not fall within the definition of "AI products" for this purpose.

3. AI-Products Liability Standard (Article 9)

In determining whether an AI product is defective, including whether it presents an unreasonable danger, courts should consider multiple factors. The SPC Explanatory Article notes that, unlike traditional products whose defects are often assessed by reference to their fixed condition at manufacture, risks associated with AI products may emerge through autonomous learning and interaction with the environment, while opacity in their decision-making may make defects more difficult to assess. Against that background, the relevant factors that courts should consider when evaluating liability include:

• The nature and intended use of the AI product — the characteristics of the AI product and the purposes for which it is designed to be used;

• Its capacity for autonomous learning — including whether risks may emerge as the AI product learns and interacts with its environment;

• Upgrades and updates — the AI product's relevant upgrade and update status in assessing the alleged danger;

• The degree of user control over the AI system — the extent to which operation of the AI system remains subject to user control; and

• Compliance with applicable national and industry standards — noncompliance with applicable safety standards may establish a defect; compliance, however, does not preclude a defect finding where the AI product nevertheless presents an unreasonable danger.

The Opinions further emphasize the importance of sufficient disclosures and warnings. Chinese courts are expected to pay particular attention to whether manufacturers and sellers of AI products have provided truthful explanations and sufficiently clear warnings regarding the AI product's intended use cases, its inherent limitations and reasonably foreseeable risks.

1. Copyright Infringement Liability among AI Developers, AI Service Providers and AI Users

Article 12 of the Opinions addresses the liability standard where AIGC infringes copyright of prior works. In determining the copyright infringement liability among AI developers, AI service providers and AI users, courts should consider the following factors:

• The type of AI service — Different AI services have different technical architectures, operating logic and risk profiles; for example, AI content-generation services may call for a higher duty of care than AI information-query services.

• The characteristics of the relevant industry — Industries differ in their reliance on and use of AI, which may affect the duties of care of AI developers and service providers.

• The source of training data — The legality of training data is an important consideration in assessing an AI developer's conduct, and the AI developer who raises a non-infringement defense may need to explain its data sources and training process.

• The level of participation by each party — The participation of AI developers, AI service providers and AI users affects their ability to control the infringement and the role each plays in generating AIGC infringing content.

• The necessary measures taken to prevent infringement — Whether AI developers or AI service providers have taken necessary measures is relevant to whether they have exercised reasonable care. If an AI developer or AI service provider fails to take precautionary measures despite knowing that an AI model may reliably generate infringing AIGC that is substantially similar to prior works, then such AI developer or AI service provider shall be found to have acted at fault.

• The economic benefits or profits obtained — Whether, and to what extent, a party profits from infringing AIGC is also relevant to the allocation of liability among all stakeholders.

Article 12 also provides that, where an AI user knows or should know about the existence of prior copyrighted works but nonetheless uses AI to generate infringing AIGC that is substantially similar to the prior works, the user should be found liable for copyright infringement, absent a statutory license, fair use or other reasonable defenses. The SPC Explanatory Article further explains that the AI user may act through prompts, parameter settings, input materials or similar means and cannot avoid liability merely by arguing that the allegedly infringing content was generated by AI.

2. Evidentiary Obligations of AI Developers Raising a Non-Infringement Defense (Article 12)

Where an AI developer raises a non-infringement defense, Article 12 of the Opinions provides that the court should order the AI developer to provide evidence concerning the source of training data, training-process records, model operating mechanisms and relevant scientific and theoretical foundations.

The SPC Explanatory Article explains that much of the information necessary to establish how an AI model operates may be unavailable to a copyright owner or other external claimant. At the same time, Article 12 provides that a right holder alleging that an AI provider used algorithmic technology to infringe copyright must provide relevant evidence. The SPC Explanatory Article emphasizes that this allocation addresses information asymmetry without reversing the basic burden-of-proof rule.

The Opinions also make clear that when AI is used to facilitate other forms of intellectual property infringement or unfair competition, including counterfeiting, false advertising, traffic inflation and fake orders, the relevant actors may be held liable under applicable law.

The rapid commercialization of technologies such as AI face-swapping, voice cloning and digital resurrection has created new forms of infringement/misappropriation involving individuals' names, likeness, voices, reputation and other personality interests.

Article 4 of the Opinions makes clear that the legal protection of personality interests extends to certain AI-generated or digitally reconstructed representations of individuals.

1. AI Infringement of Images (Article 4)

First, unauthorized AI processing of an individual's name or likeness may constitute infringement where it generates an identifiable virtual digital representation of that person and the representation is used or publicly disseminated without consent. The SPC Explanatory Article further explains that the concept of "likeness" is not limited to a person's facial features; it could also extend to that individual's clothing, body movements or other external characteristics so long as such features are sufficient for the public to identify a particular individual.

2. AI Infringement of Voices (Article 4)

Second, in terms of AI voice cloning, the Opinions explain that where the general public can identify a person based on vocal characteristics such as timbre, intonation or pronunciation style, the voice may fall within the protected scope of that individual's voice rights and interests. Article 4 itself provides that using an individual's voice as AI training material without consent and generating synthetic speech that allows the individual to be identified may constitute infringement.

Further, the Opinions address that the unauthorized use in AIGC of an individual's image and/or voice to engage in inappropriate conduct or dissemination of false statements that lower his or her social reputation may constitute infringement of reputation rights.

3. Protection of Deceased People under the AI Age (Article 4)

Article 4 of the Opinions also addresses the "digital resurrection" of deceased people. Where AI technology is used without authorization to create or exploit a deceased person's virtual digital image in a manner that infringes the deceased's name, likeness, reputation or other protected personality interests, the close relatives of such deceased person may seek legal action pursuant to Article 994 of the Civil Code.

4. AI Intrusion on Privacy (Article 5)

Article 5 of the Opinions separately addresses AI-enabled "doxxing" (i.e., using AI to track or analyze publicly available information to uncover private information about a specific individual) and other privacy intrusions. It provides that where AI is used for the purpose of prying into a specific individual's privacy, to track or to analyze publicly available information to obtain private information, to disclose or publicize the private information obtained, or to use such information to disturb the individual's private life, such conduct constitutes an infringement of the right to privacy.

Article 6 of the Opinions sets out the guidelines as to how to process publicly available personal information in AI model training.

1. Consent

In principle, where an AI developer processes, within a reasonable scope, personal information that an individual has voluntarily made public or that has otherwise been lawfully made public, and that the individual has not expressly refused such processing, such data processing for AI model training will generally not be treated as an infringement of the individual's personal information rights. However, where the processing has a major impact on the individual's rights and interests, the AI developer must obtain consent in accordance with applicable law.

The SPC Explanatory Article further explains that where an individual has "expressly refused" such processing when making the information public, or subsequently "objects to such processing" after becoming aware that the information is being used for AI model training, the individual's express refusal then defeats the legitimacy of the data processing.  In other words, for AI developers, "publicly available" personal information does not mean that it is "freely usable."

2. Within Reasonable Scope

Article 6 further provides that, in assessing whether processing remains within a "reasonable scope," courts should consider (1) the necessity and appropriateness of the processing purpose in light of the model's functions, (2) the type and sensitivity of the information and its potential impact on individual rights, (3) the context of disclosure and (4) the scope of use that the individual could reasonably expect.

Article 7 of the Opinions addresses two scenarios involving GenAI-generated infringement: content automatically generated by GenAI that infringes personality rights, and infringing content maliciously induced by users through prompts.

The SPC Explanatory Article explains that the traditional "safe harbor" or notice-and-action framework may be applied by reference to GenAI service providers: where an AI service provider receives notice of alleged infringement and promptly takes reasonable and necessary measures consistent with current technological capabilities, it generally should not bear tort liability.

The SPC Explanatory Article explains that "necessary measures" in the GenAI context may include: (1) blocking instructions/prompts that generate infringing AIGC, (2) stopping further generation of infringing content, and (3) warning, restricting or suspending services for infringing users. The SPC Explanatory Article further explains that the adequacy of such measures should be assessed under a fault-based standard, taking into account their reasonableness and timeliness in light of current technological capabilities.

The Opinions also clarify the treatment of inventions created with the assistance of AI.

Article 14 of the Opinions provides that an AI-related invention may constitute patentable subject matter where it uses technical means that follow the laws of nature, solves a technical problem, and produces a technical effect consistent with the laws of nature, unless it violates laws or social morality, harms public interests, or no natural person has made a substantive contribution.

Article 14 further provides that where a natural person uses AI to complete an invention and has made a creative contribution to the substantive features of the invention, that natural person should be recognized as the inventor. Article 14 also provides that AI-related patent specifications must satisfy the sufficient-disclosure requirement.

Article 11 of the Opinions specifically discusses the liability arising from accidents involving autonomous vehicles or vehicles equipped with assisted-driving functions, which should be determined under the Civil Code and the Road Traffic Safety Law. The SPC Explanatory Article notes that current disputes primarily involve vehicles equipped with assisted-driving functions at Level 2 or below, for which traditional traffic-accident rules continue to apply.

Where an accident results from a product defect in an autonomous or assisted-driving vehicle, the injured party may seek product liability against the manufacturer or seller. Where a defect in an assisted-driving vehicle contributes to the damage and the driver is also at fault, the injured party may seek compensation from both the driver and the vehicle manufacturer or seller.

The Opinions also address misleading commercial representations. Where manufacturers or sellers make false or misleading claims regarding a vehicle's level of automation, intelligence, performance or intended use, and consumers' lawful interests are harmed as a result, consumers may seek civil liability against the manufacturer or seller.

A noteworthy procedural development concerns access to vehicle-operation data. Article 11 permits courts, where necessary to determine accident causation, to require manufacturers, sellers or operators as data controllers to provide truthful and complete autonomous- or assisted-driving event records. The SPC Explanatory Article explains that this responds to proof difficulties where key data are held by enterprises.

[1] Supreme People’s Court, Supreme People’s Court Releases Opinions on the Trial of Cases Involving Artificial Intelligence Disputes in Accordance with the Law (Sept. 7, 2026), https://mp.weixin.qq.com/s/1wwUS-nT6veViuSp47YKkw. See also Zhou Jiahai, Si Yanli, Qin Yuanming, Jia Yuhui & Zhang Yin, Understanding and Application of the Opinions of the Supreme People’s Court on the Trial of Cases Involving Artificial Intelligence Disputes in Accordance with the Law (Sept. 7, 2026), https://mp.weixin.qq.com/s/HAoZD6HTWykItHiysz-zzQ [hereinafter SPC Explanatory Article].

[2] See Chinese AI Case Tracker, King & Wood, https://www.kingandwood.com/cn/en/insights/latest-thinking/aicasesmap.html (last visited Sept. 14, 2026) (the tracker currently includes more than 80 published AI-related cases and is being continuously updated).