Ms. Haiyu Li, Lawyer and Partner of Chofn IP
On September 7, 2026, the Supreme People's Court (SPC) of China issued the "Opinions on the Lawful Adjudication of Cases Involving Artificial Intelligence (AI) Disputes" (hereinafter referred to as the "Opinions"). This is China's first comprehensive judicial policy document systematically regulating the adjudication of AI-related disputes.
The Opinions construct the adjudicative framework for AI-related copyright infringement under Rule 12. For a long time, the crux of AI copyright disputes has not been the absence of rules, but rather the structural imbalance in the allocation of the burden of proof.
Rule 12 addresses the "technical black box" problem that has long plagued rights holders through an innovative solution: a "piercing allocation of evidentiary burden". Under this approach, where AI-generated content infringes another party's copyright, the people's courts shall reasonably determine the liability of AI developers, providers, and users in accordance with the law, after comprehensively considering factors such as the type of AI service, industrial characteristics, sources of training data, the degree of participation of each party, necessary measures taken, and profit-making circumstances.
In the past, rights holders had no knowledge of training data or model mechanisms, yet bore the burden of proving that infringement was established. Rule 12 addresses this structural dilemma by imposing special evidentiary obligations on AI developers: once a non-infringement defense is raised, they shall provide supporting evidence such as the sources of training data, records of the training process, the model's operating mode, and scientific theoretical bases. This essentially shifts the burden of proof to the party possessing technical information, so that "lack of knowledge" can no longer serve as a strategic defense to evade liability.
As for specific AI providers, where a right holder claims that they have used algorithmic technology to infringe upon its copyright, the right holder shall provide relevant evidence. In other words, the Opinions establish differentiated evidentiary rules for AI developers and providers: developers bear relatively heavier evidentiary obligations, while for providers, the burden remains primarily on the rights holders to provide evidence.
At the same time, Rule 12 explicitly establishes progressive elements of infringement for AI user liability. To be specific, infringement shall be deemed established where an AI user knows or should have known that a prior work exists uses AI to generate work substantially similar to the prior work, and has no reasonable defense grounds, such as whether the fair use defense provided in copyright law may apply. This means that when determining an AI user's tort liability, the courts will focus on the user's subjective state, for example, whether the users:
- directly entered a work title, author name, or character name in the prompt;
- asked the model to "imitate" or "replicate" a specific work or style;
- uploaded a prior work as a reference image, text, or audio;
- made multiple rounds of adjustments to approach a specific work;
- screened, modified, or commercially used the generated content; or
- had an obvious possibility of accessing the prior work.
The above layered design reflects the adjudicative logic of "matching liability with capacity for control", incorporating developers, providers, and users into a differentiated liability attribution system, abandoning the "one-size-fits-all" approach to joint liability, and better balancing the relationship between the protection of rights and interests and innovative development.
However, the Opinions fail to address two frontier issues: the copyrightability of AI-generated content and the legal characterization of using others’ works without authorization to train large AI models. This is because significant disagreements remain, and deeper understanding and further exploration are still needed before official rules can be put into place. Resolving these two questions requires the courts at all levels to ground their analysis in existing copyright law provisions and to make case-by-case determinations based on factors such as the degree of human intellectual input by AI users and the foreseeability and control of risks by AI developers and providers, thereby continuously accumulating judicial adjudication experience.
