On 7 September 2026, China’s Supreme People’s Court (SPC) issued the Opinions on the Trial of Disputes Involving Artificial Intelligence in Accordance with Law (the “Opinions”).
In recent years, discussions of AI-related IP rules have largely focused on the application and examination stage, for example, whether algorithms are eligible for patent protection and how examination standards should be applied. The newly issued Opinions instead address a more practical question from the perspective of judicial adjudication: when AI R&D, deployment and commercialization give rise to disputes, how should IP liability be allocated? And how should businesses prepare evidence, contracts and compliance systems?
Articles 12 to 16 of the Opinions specifically address AI-related IP disputes, covering in turn infringement involving AI-generated content, open-source software, AI-related inventions and patents, technology contracts, and the use of data. For businesses conducting AI R&D, licensing, technology cooperation or product deployment in China, these rules mean that IP management should no longer stop at “what rights have been filed”, but should extend across the entire lifecycle, including the provenance of training data, model development records, management of open-source components, documentation of R&D contributions, technology contract design, and preparation of litigation evidence.
I. Infringement Liability for AI-Generated Content: A Holistic Assessment of Control, Degree of Participation and Duty of Care, among Other Factors
Article 12 of the Opinions makes clear that, where AI-generated content infringes another party’s copyright, courts will comprehensively consider factors including the type of AI service, industry characteristics, the source of training data, the degree of participation of the parties, necessary measures taken, and profits obtained, and will reasonably determine the liability of developers, providers and users. This means that the mere fact that “the content was generated by AI” does not itself provide a defense. Liability will instead turn more closely on who was able to control the risk, who had a material influence on the generation of the content, and who was in a position to take preventive measures.
The Opinions also make important arrangements regarding evidence. Where an AI developer raises a non-infringement defense, the court shall order it to provide evidence concerning the sources of training data, records of the training process, the model’s mode of operation and scientific grounds. For businesses, this turns “data and model traceability” from a purely technical governance issue into a potential litigation capability issue. Training-data inventories, licenses covering data sources, model-version records, filtering measures, logs and internal decision records may all directly affect evidentiary issues and the determination of liability in future disputes.
Notably, the SPC stated in its official Q&A that the Opinions do not reach conclusions on two issues that remain highly contentious: 1) whether AI-generated content itself is eligible for copyright protection; and 2) how the unauthorized use of other parties’ works to train large AI models should be legally characterized. Businesses should therefore not read the Opinions as having resolved all copyright issues involving generative AI, and should continue to take a cautious approach to training-data authorization and output-related risks.
II. Open-Source AI: Liability of Relevant Parties — Potential Exemptions from Liability Depend on License Compliance, Risk Disclosure and Compliance Management
Article 13 is the first provision to address open-source software liability specifically from the perspective of AI disputes. When determining the liability of open-source software developers and providers, as well as downstream developers and providers, courts will consider the type of open-source license, specific restrictions on rights, safety and compliance measures, and the degree of information disclosure. Where a developer or provider makes certain code modules used for AI R&D available free of charge on an open-source basis and publicly explains their functions and safety risks, a court may find that the developer or provider is not liable if a third party’s use of the module results in infringement.
This rule is directly relevant to companies using international open-source models, code repositories and foundational components. Open source does not mean “no liability”, nor does it mean that license conditions can be ignored. Businesses should maintain a component inventory, identify attribution, copyleft, patent-license, redistribution and disclaimer requirements under different licenses, and preserve records of versions and use cases. For parties releasing open-source components, clear disclosure of functional boundaries, known risks and applicable limitations may also become an important factor in future assessments of liability.
III. AI-Assisted Inventions: The Courts Further Confirm the Core Standard of “Substantial Human Contribution”
In the patent context, Article 14 aligns relatively clearly with the direction of China’s recent patent examination policy. The Opinions state that an AI-related invention may constitute patent-eligible subject matter where it uses technical means that comply with the laws of nature, solves a technical problem and achieves a technical effect consistent with the laws of nature.
At the same time, where a natural person uses AI to complete an invention and makes a creative contribution to the substantive features of the invention, that natural person should be recognized as the inventor.
This is consistent with the general direction of current Chinese practice for AI-related patent applications: AI may serve as an R&D tool, but the inventor must still be a natural person who has made a creative contribution.
For multinational companies and R&D teams, the important issue is not simply whose name appears as inventor on the application form, but how that contribution can be demonstrated. Project initiation documents, records of prompts and parameter adjustments, experimental design, iterations of technical solutions, R&D meeting minutes and records showing how the final solution was developed may all help explain the role played by natural persons in the making of the invention.
The Opinions also reiterate the requirement for sufficient disclosure in the specification: inventions completed using AI are subject to the same sufficiency-of-disclosure requirements as inventions completed entirely by natural persons. In other words, if the patent specification for an invention completed using AI describes the technical solution to a level that enables a person skilled in the art to carry out the invention, the corresponding requirement for sufficient disclosure is satisfied. Importantly, the application documents should identify the model architecture, training methods, key parameters, input-output relationships or specific application scenarios that constitute the inventive contribution; they should not merely describe the result. “Black-box” descriptions of AI technology may not only affect grant, but may also create risks in subsequent validity and litigation proceedings.
IV. AI Technology Contracts: From “Outcome Commitments” to More Precise Provisions on the R&D Process and Reasonable Efforts
Article 15 provides that, when hearing disputes arising from AI technology development, transfer, licensing, consulting and service contracts, courts should determine liability for breach of contract in accordance with the contractual terms, while fully considering the characteristics of AI R&D and whether the technology developer has made reasonable efforts.
AI R&D is highly uncertain, and the same model and training pathway will not necessarily produce completely identical results. For parties to technology cooperation, a contract should not merely contain a broad commitment to “achieve a certain result”. Instead, it should, to the extent possible, specify who supplies the data, the training and testing conditions, evaluation metrics, deliverables, model versions, acceptance methods, iteration responsibilities, third-party components, ownership of intellectual property, and the allocation of responsibility where the expected results are not achieved.
Cross-border R&D and licensing projects should also address data transfers, confidentiality and open-source compliance. The more clearly “reasonable efforts” can be translated into verifiable R&D obligations, the more uncertainty in fact-finding can be reduced if a dispute arises.
V. Data Is Becoming a Core Asset Capable of Layered Protection
Article 16 reflects an important trend: data used in AI competition does not have to qualify as a traditional copyright subject matter before it can receive legal protection.
The Opinions make clear that courts should protect corresponding data interests where data has been lawfully obtained through collection and generation, derivative creation, assignment, licensing or other means. Data or datasets that qualify as compilation works or other copyrightable works may be protected under the Copyright Law; those that qualify as trade secrets may be protected under the Anti-Unfair Competition Law; and even where data does not constitute a trade secret, certain improper uses of data may still be regulated under the Anti-Unfair Competition Law.
Accordingly, businesses should move from asking simply “do we own copyright in the data?” to adopting a layered rights strategy. They should first confirm that the data has been lawfully obtained and that contracts clearly define ownership and permitted uses; then consider whether the selection or arrangement of a dataset is sufficiently original; and use access controls, permission management, confidentiality agreements and other measures to preserve trade-secret status. For data assets that cannot be brought within traditional exclusive-right regimes, competition law and contractual mechanisms may still provide protection.
VI. What Businesses Should Prepare in Advance: Evidence
Although the evidentiary rules appear in Part IV of the Opinions rather than in the dedicated IP section, their practical impact on AI-related IP disputes is equally important. Courts may require parties to provide full evidence concerning disputed technical facts. Where documents or electronic data are controlled by one party and that party refuses to produce them without justification, the court may make findings adverse to that party. Where AI-generated content is submitted as evidence of infringement, courts will also consider the design of prompts, the effect of prompts on the output, the degree of similarity between the generated content and the work in which rights are asserted, the consistency of repeated testing, as well as factors such as model training, algorithm design and content-filtering mechanisms.
This means that businesses should move “litigation readiness” forward into the R&D and commercialization stages. Model versions, training data, prompts, test results, source-code commit records, third-party licenses, contractual correspondence and access logs should, where possible, form a traceable, searchable and verifiable evidentiary chain. Many AI-related IP disputes may ultimately be lost not because the legal rules are unfavorable, but because critical facts cannot be proved.
Conclusion
The significance of the SPC Opinions lies in moving AI-related IP issues beyond the single question of “application and grant” and into the full lifecycle of R&D, deployment, contracts, data management and dispute resolution.
For overseas applicants and businesses conducting AI-related activities in China, China IP strategy for AI should evolve accordingly: patent filings remain important, but a patent portfolio alone is no longer sufficient to address all risks arising from AI commercialization.
A more resilient approach is to manage inventorship contributions, training-data provenance and open-source components from the R&D stage; define technical objectives, IP ownership and data boundaries in transactions; continuously assess risks relating to generated content and competition law during commercialization; and preserve technical and business records capable of withstanding judicial scrutiny throughout the process. In this way, intellectual property can be transformed from a static collection of rights into enforceable assets that support AI innovation and commercialization.
If you have any questions regarding patents, trademarks, litigation, enforcement or other intellectual property matters, please feel free to contact us. For patent or trademark matters, please contact info@anxinfonda.com, Tel: +86-10-8273-0790. For litigation or legal matters, please contact info@bhtdlaw.com, Tel: +86-10-8273-7958.
