China's Supreme Court Issues Landmark Judicial Rules on AI: Implications for Product Liability, IP, Data, and Generative AI 

The Opinions set out three principles for adjudicating AI-related disputes—putting people first, fostering innovation-driven development, and safeguarding a safety baseline—and establish the general fault principle for AI-related liability unless specific laws require strict liability (e.g., Product Quality Law) or presumed negligence (e.g., PIPL).

AI PRODUCT LIABILITY

AI-related product liability claims may arise from (i) inherent product defects, (ii) intentional torts or user/driver negligence, (iii) false or misleading promotional statements, and (iv) failure to warn or inadequate warnings and instructions. Manufacturers, distributors, and users/drivers may bear separate or joint liability under the general tort provisions of the Civil Code, the Opinions, and other applicable laws.

Article IX addresses product liability for AI-enabled physical products under the Product Quality Law. Where a defect in an AI product causes damage, the producer and seller bear product liability under existing law. The Opinions establish a multifactor test for assessing whether an AI product poses an "unreasonable danger" and thus constitutes a defective product:

• The nature and purpose of the AI product;
• The product's self-learning ability;
• The product's upgrade and update status;
• The user's degree of control over the system; and
• Whether the product complies with relevant national and industry standards.

The Opinions place special emphasis on whether the producer and seller provided truthful explanations and clear warnings regarding applicable scenarios, inherent limitations, and foreseeable risks. Full disclosure may be critical to determining product defects and liability.

Autonomous vehicles ("AVs") are AI-enabled products subject to the Product Quality Law and Article IX of the Opinions. Article XI further addresses several AV-specific scenarios:

• Combined defect and driver fault: Where vehicle defects and driver fault combine to cause the same damage, courts shall support claims against both the driver and the manufacturer/seller under Article 1172 of the Civil Code (concurrent liability).
• False or misleading claims: Where manufacturers or sellers misrepresent the automation level, intelligence, or performance of autonomous or driver-assist vehicles, courts shall support consumer civil liability claims.
• Data production orders: Courts may compel manufacturers, sellers, operators, and other data controllers to produce autonomous driving and driver-assist event records and other data necessary to ascertain the facts of a case.

In Shen v. Taizhou Automobile Sales and Service Company (2023), the Taizhou Intermediate People's Court dismissed the plaintiff's product liability claims, holding that an Automatic Emergency Brake ("AEB") system's inability to detect lateral vehicles was a technological limitation rather than an "unreasonable danger," and that the user manual clearly disclosed the system's constraints. The driver was held liable because continuous acceleration overrode the AEB by design, causing the accident.

Practical Implications

• Strict liability applies to manufacturers and sellers of hardware-integrated AI products when a defect poses an unreasonable danger to personal or property safety. Compliance with national and industry standards is a key evaluation factor but is not a safe harbor.
• Companies manufacturing or selling AI-enabled physical products (e.g., robotics, smart medical devices, AI-powered industrial equipment) in China should review product disclosures, warning labels, and user documentation to minimize liability. Clear disclosures of inherent limitations (e.g., AI output unpredictability) and intended-use scenarios should be prominently displayed in user manuals, onboarding screens, and disclaimers.
• The self-learning capability of AI products introduces a dynamic element to defect analysis. Products may evolve post-sale through updates and learning, and courts will consider this evolution in assessing liability.
• Companies should exercise caution in advertising to avoid overstating product capabilities.
• The data production power is particularly significant: courts can compel disclosure of driving event logs and system data, creating both litigation risk and regulatory exposure for AV manufacturers. Companies should maintain robust data retention policies for autonomous driving records.

AI-RELATED COPYRIGHT INFRINGEMENT

Article XII provides a comprehensive framework for allocating liability when AI-generated content infringes copyrights. Courts are directed to consider a range of factors in determining responsibility among AI developers, providers, and users:

• The type of AI service and industry characteristics;
• The source of training data;
• The participation and conduct of all parties (developer, provider, user);
• Necessary measures taken to prevent infringement; and
• Profits gained from the infringing activity.

The Opinions impose specific evidentiary burdens:

• Developers asserting non-infringement must provide evidence regarding the source of training data, training process records, model operation mode, and scientific theoretical basis.
• Users who know or should know of a prior work and generate substantially similar content through AI also bear infringement liability.

The Opinions seem to avoid addressing whether AI-generated works are themselves copyrightable, leaving this significant issue unresolved at the national level, despite the Beijing Internet Court's prior ruling in Li v. Liu (2025) that output from the AI model Stable Diffusion, including detailed images generated from text descriptions, could be copyrightable.

In the Xinchuanghua case (2024), the Hangzhou Internet Court held the platform contributorily liable for derivative copyright violations where the platform provided image generation tools and contributed to users' copyright infringement. It also established a heightened "duty of care" requiring platforms to implement filters, warnings, and watermarks for well-known works.

Practical Implications

• AI developers should maintain detailed records of training data provenance, the training process, and model architecture; these records may be compelled in litigation.
• The allocation of liability among developers, providers, and users creates a shared responsibility model. Contractual indemnification provisions among these parties will be critically important.
• The "know or should know" standard for users means companies using AI tools to generate content should implement screening processes to identify potential similarity with existing copyrighted works.

AI-RELATED PATENT RULES

Article XIV addresses patent eligibility and inventorship for AI-related inventions. The Opinions provide three significant rules:

• Patentable subject matter: Where an AI-related invention employs technical means, solves a technical problem, and achieves technical effects, courts shall recognize it as patentable subject matter subject to the standard exclusions for illegality, immorality, harm to the public interest, or where no substantial human contribution was made.
• Human inventorship: A natural person who uses AI to complete an invention and makes a creative contribution to the substantive features of the invention shall be recognized as the inventor. This reaffirms that only natural persons can be inventors under Chinese patent law, while clarifying that AI-assisted inventions are protectable.
• Sufficiency of disclosure: A patent specification describing an AI technical solution at a level sufficient to enable a person having ordinary skill in the art to implement the invention meets the sufficiency of disclosure requirement.

Practical Implications

• Companies filing AI-related patents in China should document the human inventor's creative contribution to the substantive features of the invention to satisfy the inventorship requirement.
• The "substantial human contribution" requirement for patentability means fully autonomous AI inventions without meaningful human involvement may not be patentable in China.

OPEN-SOURCE SOFTWARE LIABILITY

Article XIII addresses tort liability for open-source software used in AI development and takes a lenient approach to encourage open-source providers. When adjudicating cases involving open-source software, courts must consider:

• The type of open-source license;
• The specific restrictions on rights;
• Security compliance measures; and
• The degree of information disclosure.

Importantly, if an open-source software developer or provider provides free, open-source code modules for AI research and development and publicly explains the code's functions and security risks, the court may find the developer or provider not liable for tort when third parties infringe using those modules.

Practical Implications

• This lenient approach toward open-source providers reflects China's overall strategy of encouraging open-source AI development. It provides a degree of safe harbor for contributors who make good-faith disclosures about their open-source code or modules.
• Open-source AI developers should ensure comprehensive documentation of functions, known limitations, and security risks to maximize the benefit of this liability exemption.
• Downstream developers incorporating open-source AI components should conduct thorough due diligence on the license terms, disclosed risks, and security posture of the open-source modules they adopt.

GENERATIVE AI TORT LIABILITY FOR INFRINGEMENT OF PERSONALITY RIGHTS

Article VII establishes a notice-and-takedown framework for generative AI tort liability involving personality rights. The framework distinguishes two liability scenarios:

• Provider Liability: "Safe harbor rule" explicitly applies (Article 1195 of the Civil Code). If AI-generated content infringes reputation, privacy, or other personality interests, and the provider fails to take timely measures after receiving notice, the provider bears tort liability for the resulting harm. A valid notice must include (i) preliminary evidence of infringement and (ii) the rights holder's identity. The "red flag rule" remains judicially applicable (Article 1197 of the Civil Code) and providers will be liable if they knew or should have known of user infringements.

• User Liability: A user who maliciously induces generative AI to produce infringing content through infringing prompts bears independent tort liability. Critically, if the rights holder notifies the provider and the provider fails to take timely measures, the rights holder may pursue claims against the user and/or the provider.

Practical Implications

• Generative AI providers operating in China should implement notice-and-takedown mechanisms that can stop infringing generation and block specific prompts.
• The preliminary evidence and identity requirements establish a threshold before the provider must act and offer some protection against frivolous complaints.
• The extension of the safe harbor rule to generative AI is noteworthy: it treats AI providers analogously to internet service providers and applies the established notice-and-takedown framework to this new technology.

BIG DATA PRICE DISCRIMINATION

Article X addresses algorithmic price discrimination―when a business operator uses algorithms to impose unreasonable differential treatment on transaction prices or conditions for the same product or service. Courts must consider:

• Whether the differential treatment substantially restricts or damages consumers' right to know, right to choose independently, and right to fair trade;
• Whether transaction conditions are based on consumers' consumption preferences, willingness to pay, ability to pay, browsing history, and other personal information targeting them individually;
• Whether the practice violates the principle of good faith and business ethics; and
• Whether the reasons for differential treatment are justified, sufficient, and non-discriminatory.

In a 2021 case, the Shaoxing Intermediate People's Court found that a leading online travel platform violated Article X by failing to distinguish self-operated from agency-channel listings, using personal data for price targeting without adequate disclosure, and neglecting to monitor agency pricing despite having the authority and capability to do so. The court rejected the platform's defense that a third-party agent set the price, holding it bore a statutory duty to disclose such arrangements. Treble damages were upheld under the Consumer Rights Protection Law.

Practical Implications

• E-commerce platforms and any business using algorithmic pricing in China should audit their pricing algorithms for differential treatment that could be characterized as discriminatory and comply with relevant mandatory disclosure requirements regarding the use of algorithmic pricing.
• The emphasis on personal data targeting (browsing history, willingness to pay) aligns with broader PIPL compliance and suggests that personalized pricing models carry significant legal risk.

AI DATA PROTECTION

Article XVI provides a multilayered framework for protecting data rights in the AI context, leveraging several existing laws:

• Data rights of developers: Courts shall protect the data rights of AI developers who legally acquire data through collection, generation, derivative creation, transfer, or licensing.
• Compilations as copyrightable works: Data compilations or datasets that meet the threshold of originality are protected under the Copyright Law.
• Trade secrets protection: Data qualifying as trade secrets are protected under Article 10 of the Anti-Unfair Competition Law, enumerating different types of infringement of trade secrets.
• Non-trade-secret data: Data not qualifying as trade secrets are protected under Article 13 of the Anti-Unfair Competition Law, a catch-all clause designed to cover internet and AI-related activities such as using data, algorithms, technology, or platform rules to interfere with or disrupt other companies' lawful online products or services, or obtaining or using other companies' lawfully held data through fraud, coercion, or circumventing technical measures.
• Antitrust liability: Using data, algorithms, or other technical means to reach monopoly agreements or abuse market dominance falls under the Anti-Monopoly Law.
• Data integrity and security: Fabricating interference data, maliciously labeling data, or adversarial sample attacks that damage AI operational security may fall under Article 13 of the Anti-Unfair Competition Law.

Practical Implications

• AI companies should assess their data assets across all three protection layers (copyright, trade secrets, and unfair competition) and implement appropriate safeguards.
• The recognition of legal data acquisition rights is a positive signal for the AI industry, but companies must ensure they have a clear legal basis for all data used in AI training and operations.

CONCLUSION

The Opinions represent a significant milestone in China's regulation of AI, setting out legal principles for a wide range of AI-related issues.

AI compliance is critical for global companies seeking to minimize AI-related risks and litigation exposure. While China, the United States, the European Union, and other jurisdictions are converging on the need for explainable, well-documented, secure, and accountable AI systems, they are not converging on a single legal architecture. Risk classification frameworks, operational expectations, enforcement mechanisms, and litigation risks vary significantly across regimes. A practical cross-border compliance strategy should therefore begin with a foundation of common controls—including data provenance, technical documentation, human oversight, user transparency disclosures, cybersecurity by design, vendor contractual safeguards, and antitrust review—and then layer jurisdiction-specific legal analysis on top.