This is an informal case summary prepared for the purposes of facilitating exchange during the 2026 WIPO IP Judges Forum.
Session 1: IP and Artificial Intelligence (AI)
Supreme People’s Court of China [2025]: Company A v Company B (2023) ZGFZMZ No. 1503
Date of judgment: August 22, 2025
Issuing authority: Supreme People’s Court of China
Level of the issuing authority: Final Instance
Type of procedure: Judicial (Civil)
Subject matter: Undisclosed Information (Trade Secrets)
Plaintiff/Appellant: Company A
Defendant/Respondent: Company B; Zhang; Li; Wang; Huang
Keywords: Trade secrets; Artificial intelligence (AI); Algorithmic technology; Burden of proof
Basic facts: Company A was dedicated to developing a fingertip recognition and point-to-read English word product called “Certain English Reading Companion”, which enables users to point at text in books with their fingers for rapid recognition and translation. Company A claimed that it had formed relevant trade secrets before April 2019, including algorithms comprised of fingertip recognition-related code and an image database. Zhang was formerly one of Company A’s shareholders and served as Company A’s Chief Technology Officer. Li, Wang, and Huang were also three former employees of Company A, all engaged in technology development related to the "fingertip recognition" project and had access to the relevant technical information. Zhang left Company A in March 2019 and established Company B in late May of the same year. Li, Wang, and Huang subsequently left Company A and joined Company B, becoming shareholders or key technical personnel. Company A alleged that Zhang, Li, Wang, and Huang, without permission, disclosed and allowed Company B to use Company A's trade secrets, and that Company B utilized these trade secrets to provide technical support to other companies that subsequently launched products with fingertip recognition and point-to-read word functions. Accordingly, Company A filed a lawsuit requesting that Company B, Zhang, Li, Wang, and Huang immediately cease misappropriation of trade secrets and jointly compensate Company A for economic losses (including reasonable expenses) of RMB 1 million.
The first-instance court held that the evidence provided by Company A could not reasonably indicate that its asserted technical information had been misappropriated. Accordingly, the first-instance court rendered a judgment dismissing Company A's claims. Dissatisfied, Company A appealed to the Supreme People's Court.
Held: At the end of 2025, the Intellectual Property Court of the Supreme People's Court of China concluded the appeal case filed by Company A against Company B, Zhang, Wang, Li, and Huang (Hereinafter the five defendants) for misappropriation of trade secrets. he Court reversed the original judgment and held that the five defendants had jointly misappropriated Company A's trade secrets and should bear corresponding liability. This case is the first trade secrets dispute adjudicated by the Court involving artificial intelligence and algorithmic technology in the field of visual recognition. It provides an excellent interpretation of how to apply Paragraph 2 of Article 32 of the Anti-Unfair Competition Law of China (as amended in 2019) regarding the shift of burden of proof in this field.
The Supreme People's Court held on second instance that the evidence provided by Company A had reasonably indicated that the trade secrets had been misappropriated by the five defendants. First, on Company B's homepage dated October 14, 2019, Company A's "Certain English Reading Companion" product was explicitly introduced as the main product of the "Desktop Interaction Technology Platform," and the website emphasized that the product utilized AI algorithm-based fingertip positioning technology.
Second, in a prior case where Company A sued Company B based on substantially the same facts and subsequently withdrew the lawsuit, according to Company B's defense in that prior case, the accused infringing product and Company A's "Certain English Reading Companion" product both used "finger recognition and tracking" technology.
Third, when comparing demonstrations of the accused infringing product and Company A's "Certain English Reading Companion" product, with and without finger-assisted testing scenarios, both products demonstrated almost identical recognition, output, and spelling capabilities.
Fourth, from Company B's establishment (May 21, 2019) to the date when its partner's product displayed "from nothing to something" with finger positioning and recognition functions (July 4, 2019), the duration was less than two months.
Finally, Zhang, Wang, Li, and Huang were all members of the fingertip recognition project team while employed at Company A, and all had opportunities to access the relevant technical information of the asserted trade secrets.
Relevant holdings in relation to IP and artificial intelligence: The evidence submitted by the five defendants was insufficient to prove that they had not misappropriated Company A's trade secrets.
Firstly, according to Company B's claims, the source of the accused infringing technology was primarily open-source code collected from various open-source channels. However, how to integrate open-source code obtained from different channels to form an initial model and further develop a fully functional model through extensive subsequent training is both a critical and core difficulty in research and development (R&D). The process of obtaining open-source code related to fingernail recognition technology from different open-source channels to launching a commercially viable product undoubtedly requires substantial adaptation, integration, and refinement work. Taking the "fingertip recognition technology" that Company A had asserted as a trade secret as an example, the R&D process from conceptual discussion to final product launch lasted more than one year and five months. In contrast, Company B, from its establishment to providing technology enabling "finger positioning and recognition" functions to its commercial partners, took less than two months. Developing and launching an artificial intelligence visual recognition and positioning product from scratch in such a short R&D cycle or at such rapid progress is contrary to common sense and general experience.
Secondly, whether the fingertip recognition technology that Company A sought to protect as a trade secret or Company B's accused infringing technology, both inevitably involve processes such as finger recognition, finger positioning, identifying text pointed at by fingers, correctly understanding text content, and responding accordingly. This series of "actions" is actually completed with artificial intelligence participation and enabling an "artificial intelligence" product to possess human-like visual perception, recognition, understanding, and response capabilities requires high-frequency and high-intensity training of the "artificial intelligence" model beforehand, such as training how to recognize with fingers. Without data "feeding," the artificial intelligence model cannot function effectively. The richer, larger-scale, and higher-quality the data "fed" to the artificial intelligence model, the stronger its self-learning and generalization capabilities it will obtain, and the better its demonstrated cognitive level and processing ability. Without sufficient "feeding" data to facilitate reinforcement learning training of the artificial intelligence model, it is clearly unreasonable that Company B, relying solely on open-source code obtained from different open-source channels, could (a) assist its commercial partner Youmoumou in launching a new product with "finger recognition and point-to-read" functions different from Youmoumou's original product functions before the collaboration, and (b) claim the technology is "completely different" from Company A's "fingertip recognition and positioning" technology—namely a "fingernail" technology—within such a short time.
Thirdly, generally speaking, the knowledge and capabilities of artificial intelligence models originate from their training data. The training process involves teaching the model to map input data to specific labels, and the model's capabilities are strictly limited by the task boundaries defined by the training data. Artificial intelligence models do not spontaneously acquire the ability to recognize untrained categories. The quality of model output fundamentally depends on the type, scale, and quality of training data input into the model. According to the comparison demonstration video of Company A's product and the accused infringing product when simultaneously facing the same tester and the same test material, combined with facts further ascertained on second instance, in the confidential fingernail test scenario, the accused infringing product utilizing technology provided by Company B could still smoothly recognize, trigger positioning, and correctly spell and output relevant words. This fact precisely indicates that Company B's claim that the accused infringing technology adopted "fingernail recognition" technology lacks persuasiveness.
Finally, the "Certain English Reading Companion" is a cylindrical product with a camera device, and objectively there is a possibility that users may arbitrarily move and place this cylindrical product during use; therefore, the product's background needs to adjust the angle to collect samples. The accused infringing product, however, is a tablet-type product whose usage characteristics basically do not require background angle adjustment, and Company B also stated that its technology has no angle adjustment function. However, according to its product test results, the accused infringing product could also recognize words in angle-adjusted situations (i.e., scenarios where the tablet and user are at angles deviating from normal usage).
Based on the above analysis, the Court held that the five defendants had misappropriated the trade secrets. Accordingly, the first-instance judgment was reversed, and the five defendants were ordered to cease the tort action of trade secrets, and bear joint liability for RMB 500,000 in damages.
This case provides relatively in-depth analysis and reasoning on how to apply Paragraph 2 of Article 32 of the Anti-Unfair Competition Law (as amended in 2019) and how to allocate the burden of proof in trade secret disputes, providing certain reference for the adjudication of artificial intelligence and algorithm-related trade secret cases.
Relevant legislation: Law Against Unfair Competition of the People's Republic of China (CN409)