China: Supreme Court Clarifies Burden of Proof Shifting in AI and Algorithm Trade Secret Infringement Cases
Published 29 December 2025
Yu Du
On 25 December 2025, the Intellectual Property Tribunal of the Supreme People’s Court (SPC) published on its official website a trade secret infringement case it concluded on 22 August 2025. The publication aims to illustrate how the burden of proof shifting mechanism under the 2019 amendment to the Anti-Unfair Competition Law should be applied in disputes involving artificial intelligence and algorithm-related technical secrets.
The case, numbered (2023) SPC IP Civil Final No. 1503, is the first appellate judgment by the SPC concerning trade secret infringement in the field of visual recognition AI algorithms. It provides important judicial guidance on evidentiary standards and judicial reasoning in cases where the alleged infringement involves complex algorithmic systems, machine learning models, and training data that are inherently difficult for right holders to directly access or prove.
Case History
The appellant, Company A, developed an “English Reading Companion” product that enables users to point at printed text with a finger and instantly obtain word recognition and translation. Company A claimed that, prior to April 2019, it had already formed protectable technical trade secrets, including the core algorithms embodied in source code for finger-tip recognition and a proprietary image database used to train its artificial intelligence model.
One of the defendants, Zhang, was previously a shareholder of Company A and served as its Chief Technology Officer. The other individual defendants, Li, Wang, and Huang, were former employees who participated in the finger recognition project and had access to the relevant technical information. In March 2019, Zhang resigned from Company A and, in May 2019, established Company B. Shortly thereafter, Li, Wang, and Huang also left Company A and joined Company B as shareholders or key technical personnel.
Company A alleged that the individual defendants, without authorization, disclosed and allowed the use of its trade secrets by Company B, which subsequently used those trade secrets to provide technical support to third-party companies. These third parties then launched products incorporating finger recognition and point-and-read functions. On this basis, Company A filed a lawsuit in April 2021, seeking injunctive relief and joint and several damages in the amount of RMB 1 million.
The court of first instance rendered its judgment on 27 April 2023, dismissing Company A’s claims on the ground that the evidence presented was insufficient to reasonably prove that its technical information had been infringed. Dissatisfied with the judgment, Company A filed an appeal in August 2023, bringing the case before the SPC.
Court’s Reasoning and Ruling
1. Prima Facie Showing of Trade Secret Infringement
Upon appellate review, the SPC held that Company A had provided sufficient evidence to reasonably indicate that its trade secrets had been infringed, thereby meeting the threshold for applying the burden of proof shifting mechanism under Article 32(2) of the Anti-Unfair Competition Law. The Court emphasized that, at this stage, Company A was not required to prove infringement conclusively, but only to demonstrate a reasonable likelihood of misappropriation.
The Court noted that Company B’s official website had expressly introduced Company A’s product as a core reference for its “desktop interactive technology platform” and highlighted the use of AI-based finger positioning algorithms. In prior litigation involving substantially the same facts, Company B also acknowledged that both products employed finger recognition and tracking technology.
In addition, comparative demonstrations showed that, under identical testing conditions, the two products exhibited nearly identical recognition, output, and pronunciation performance. The Court further considered the extremely short development period between Company B’s establishment and the commercial launch of finger recognition functionality, as well as the defendants’ prior access to the alleged trade secrets through their employment at Company A. Taken together, these circumstances were sufficient to establish a prima facie case of trade secret infringement.
2. Defendants’ Failure to Rebut the Presumption
After the burden of proof had shifted, the defendants asserted that the accused technology was independently developed based on open-source code collected from various channels. The SPC found that this explanation failed to adequately account for the origin of the accused technology.
The Court observed that collecting open-source code does not, by itself, result in a commercially viable artificial intelligence system. Integrating code from different sources, adapting it to specific application scenarios, and training a functional AI model are among the most technically challenging aspects of AI development. By comparison, Company A’s development of its finger recognition technology took more than one year, while Company B claimed to have achieved similar results in less than two months, a timeline the Court found inconsistent with ordinary technical experience.
The Court further emphasized that artificial intelligence models rely heavily on large-scale, high-quality training data. Without sufficient data “feeding” and repeated training, an AI model cannot reliably perform complex tasks such as finger recognition, text identification, and accurate output generation. Comparative testing demonstrated that the accused product could still function when fingernails were concealed and could adapt to changes in viewing angle, directly contradicting the defendants’ claim that their system relied on a fundamentally different technical approach. Accordingly, the Court held that the defendants failed to rebut the presumption of infringement.
3. Final Judgment
Based on a comprehensive assessment of the evidence and technical circumstances, the SPC reversed the first-instance judgment and concluded that Company B and the four individual defendants had jointly infringed Company A’s trade secrets.
The Court ordered the defendants to immediately cease the disclosure, use, and authorization of use of the trade secrets, to destroy all carriers containing the relevant technical information, and to bear joint and several liability for damages in the amount of RMB 500,000.
The judgment reaffirmed the proper application of the burden of proof shifting mechanism in trade secret disputes involving artificial intelligence and algorithmic technologies, and clarified the evidentiary standards applicable to such cases.
Significance and Practical Implications
This case represents a landmark decision in the judicial protection of AI and algorithm-related trade secrets in China. It demonstrates that courts are prepared to look beyond the surface complexity of artificial intelligence systems and assess infringement through indirect but technically grounded evidence, such as development timelines, functional equivalence, personnel overlap, and the logic of machine learning training processes.
More importantly, the judgment provides concrete guidance on how burden of proof shifting should operate in technical secret cases. It confirms that a right holder is not required to directly access or compare source code to establish infringement. Instead, once the plaintiff demonstrates a reasonable likelihood of infringement, the evidentiary burden shifts to the defendant to prove independent development or lawful origin. This approach significantly reduces the otherwise insurmountable evidentiary barriers faced by trade secret owners in AI-related disputes.
Comment
The Court’s reasoning reflects an increasingly sophisticated judicial understanding of artificial intelligence, particularly the role of training data, the limits of model generalization, and the practical realities of AI product development. For rights holders, the judgment illustrates an effective litigation strategy: focusing on circumstantial but technically persuasive evidence to establish a prima facie case and activate burden of proof shifting. In doing so, the SPC has provided a practical and enforceable framework for protecting AI and algorithmic trade secrets, reinforcing confidence in judicial remedies for innovation-driven enterprises.
The case, numbered (2023) SPC IP Civil Final No. 1503, is the first appellate judgment by the SPC concerning trade secret infringement in the field of visual recognition AI algorithms. It provides important judicial guidance on evidentiary standards and judicial reasoning in cases where the alleged infringement involves complex algorithmic systems, machine learning models, and training data that are inherently difficult for right holders to directly access or prove.
Case History
The appellant, Company A, developed an “English Reading Companion” product that enables users to point at printed text with a finger and instantly obtain word recognition and translation. Company A claimed that, prior to April 2019, it had already formed protectable technical trade secrets, including the core algorithms embodied in source code for finger-tip recognition and a proprietary image database used to train its artificial intelligence model.
One of the defendants, Zhang, was previously a shareholder of Company A and served as its Chief Technology Officer. The other individual defendants, Li, Wang, and Huang, were former employees who participated in the finger recognition project and had access to the relevant technical information. In March 2019, Zhang resigned from Company A and, in May 2019, established Company B. Shortly thereafter, Li, Wang, and Huang also left Company A and joined Company B as shareholders or key technical personnel.
Company A alleged that the individual defendants, without authorization, disclosed and allowed the use of its trade secrets by Company B, which subsequently used those trade secrets to provide technical support to third-party companies. These third parties then launched products incorporating finger recognition and point-and-read functions. On this basis, Company A filed a lawsuit in April 2021, seeking injunctive relief and joint and several damages in the amount of RMB 1 million.
The court of first instance rendered its judgment on 27 April 2023, dismissing Company A’s claims on the ground that the evidence presented was insufficient to reasonably prove that its technical information had been infringed. Dissatisfied with the judgment, Company A filed an appeal in August 2023, bringing the case before the SPC.
Court’s Reasoning and Ruling
1. Prima Facie Showing of Trade Secret Infringement
Upon appellate review, the SPC held that Company A had provided sufficient evidence to reasonably indicate that its trade secrets had been infringed, thereby meeting the threshold for applying the burden of proof shifting mechanism under Article 32(2) of the Anti-Unfair Competition Law. The Court emphasized that, at this stage, Company A was not required to prove infringement conclusively, but only to demonstrate a reasonable likelihood of misappropriation.
The Court noted that Company B’s official website had expressly introduced Company A’s product as a core reference for its “desktop interactive technology platform” and highlighted the use of AI-based finger positioning algorithms. In prior litigation involving substantially the same facts, Company B also acknowledged that both products employed finger recognition and tracking technology.
In addition, comparative demonstrations showed that, under identical testing conditions, the two products exhibited nearly identical recognition, output, and pronunciation performance. The Court further considered the extremely short development period between Company B’s establishment and the commercial launch of finger recognition functionality, as well as the defendants’ prior access to the alleged trade secrets through their employment at Company A. Taken together, these circumstances were sufficient to establish a prima facie case of trade secret infringement.
2. Defendants’ Failure to Rebut the Presumption
After the burden of proof had shifted, the defendants asserted that the accused technology was independently developed based on open-source code collected from various channels. The SPC found that this explanation failed to adequately account for the origin of the accused technology.
The Court observed that collecting open-source code does not, by itself, result in a commercially viable artificial intelligence system. Integrating code from different sources, adapting it to specific application scenarios, and training a functional AI model are among the most technically challenging aspects of AI development. By comparison, Company A’s development of its finger recognition technology took more than one year, while Company B claimed to have achieved similar results in less than two months, a timeline the Court found inconsistent with ordinary technical experience.
The Court further emphasized that artificial intelligence models rely heavily on large-scale, high-quality training data. Without sufficient data “feeding” and repeated training, an AI model cannot reliably perform complex tasks such as finger recognition, text identification, and accurate output generation. Comparative testing demonstrated that the accused product could still function when fingernails were concealed and could adapt to changes in viewing angle, directly contradicting the defendants’ claim that their system relied on a fundamentally different technical approach. Accordingly, the Court held that the defendants failed to rebut the presumption of infringement.
3. Final Judgment
Based on a comprehensive assessment of the evidence and technical circumstances, the SPC reversed the first-instance judgment and concluded that Company B and the four individual defendants had jointly infringed Company A’s trade secrets.
The Court ordered the defendants to immediately cease the disclosure, use, and authorization of use of the trade secrets, to destroy all carriers containing the relevant technical information, and to bear joint and several liability for damages in the amount of RMB 500,000.
The judgment reaffirmed the proper application of the burden of proof shifting mechanism in trade secret disputes involving artificial intelligence and algorithmic technologies, and clarified the evidentiary standards applicable to such cases.
Significance and Practical Implications
This case represents a landmark decision in the judicial protection of AI and algorithm-related trade secrets in China. It demonstrates that courts are prepared to look beyond the surface complexity of artificial intelligence systems and assess infringement through indirect but technically grounded evidence, such as development timelines, functional equivalence, personnel overlap, and the logic of machine learning training processes.
More importantly, the judgment provides concrete guidance on how burden of proof shifting should operate in technical secret cases. It confirms that a right holder is not required to directly access or compare source code to establish infringement. Instead, once the plaintiff demonstrates a reasonable likelihood of infringement, the evidentiary burden shifts to the defendant to prove independent development or lawful origin. This approach significantly reduces the otherwise insurmountable evidentiary barriers faced by trade secret owners in AI-related disputes.
Comment
The Court’s reasoning reflects an increasingly sophisticated judicial understanding of artificial intelligence, particularly the role of training data, the limits of model generalization, and the practical realities of AI product development. For rights holders, the judgment illustrates an effective litigation strategy: focusing on circumstantial but technically persuasive evidence to establish a prima facie case and activate burden of proof shifting. In doing so, the SPC has provided a practical and enforceable framework for protecting AI and algorithmic trade secrets, reinforcing confidence in judicial remedies for innovation-driven enterprises.