*AI-generated translation, for reference only.

Recently, the Intellectual Property Court of the Supreme People's Court concluded the case of dispute over misappropriation of technical secrets between the appellant, A Company, and the appellees, B Company, Zhang [REDACTED], Wang [REDACTED], Li [REDACTED], and Huang [REDACTED]. The Court amended the original judgment and determined that B Company and the other four alleged infringers had committed joint infringement against the technical secrets at issue of A Company and should bear corresponding infringement liability. This case is the first dispute over misappropriation of technical secrets involving artificial intelligence and algorithm technologies in the field of visual recognition concluded by the Court, and provides a sound interpretation of how Article 32, Paragraph 2 of the Anti-Unfair Competition Law, as amended in 2019, concerning the shifting of burden of proof, should be applied in this field.

A Company was committed to developing a product named "[REDACTED] English Reading Companion," which features fingertip recognition and word point-reading, namely, recognizing and translating text in a book quickly when the user points at the text with a finger. A Company claimed that, before April 2019, it had already formed relevant trade secrets, including algorithms composed of code related to fingertip recognition technology and an image database. Zhang [REDACTED] was once one of the shareholders of A Company and served as its CTO. Li [REDACTED], Wang [REDACTED], and Huang [REDACTED] were also former employees of A Company, all engaged in technical development work related to the "fingertip recognition" project and had access to the technical information at issue. In March 2019, Zhang [REDACTED] resigned from A Company and established B Company in late May of the same year. Thereafter, Li [REDACTED], Wang [REDACTED], and Huang [REDACTED] successively resigned from A Company and joined B Company, becoming shareholders or key technical personnel of B Company. A Company claimed that Zhang [REDACTED], Li [REDACTED], Wang [REDACTED], and Huang [REDACTED], without authorization, disclosed A Company's trade secrets at issue to B Company and permitted B Company to use them, and that B Company used such trade secrets to provide technical support to other companies, which then launched products featuring fingertip recognition and word point-reading functions. Accordingly, A Company filed a lawsuit with the court, requesting an order that B Company, Zhang [REDACTED], Li [REDACTED], Wang [REDACTED], and Huang [REDACTED] immediately cease misappropriating the trade secrets at issue and jointly and severally compensate A Company for economic losses, including reasonable expenses, in the amount of RMB 1 million.

Upon trial, the Court of First Instance held that the evidence submitted by A Company could not reasonably indicate that the technical information claimed by A Company had been misappropriated. Accordingly, the Court of First Instance rendered a judgment dismissing A Company's claims. Dissatisfied with the judgment, A Company appealed to the Supreme People's Court.

Upon second instance trial, the Supreme People's Court held that the evidence submitted by A Company had reasonably indicated that the trade secrets at issue had been misappropriated by the five alleged infringers. First, on B Company's company homepage dated October 14, 2019, B Company expressly presented A Company's "[REDACTED] English Reading Companion" product as a principal product of its "desktop interaction technology platform," and emphasized that the product used AI algorithm-based fingertip positioning technology. Second, in the case in which A Company first sued B Company based on facts substantially identical to those in this case and later withdrew the lawsuit, according to B Company's defense, both the accused infringing product and A Company's "[REDACTED] English Reading Companion" product used "finger recognition and tracking" technology. Third, in a comparative demonstration between the accused infringing product and A Company's "[REDACTED] English Reading Companion" product, under testing scenarios both with and without the aid of a finger, the recognition, output, and spelling abilities displayed by the two products were almost identical. Fourth, from the establishment of B Company on May 21, 2019 to the time when the product of its business partner was shown to have acquired, "from scratch," finger positioning and recognition functions on July 4, 2019, less than two months had elapsed. Finally, when Zhang [REDACTED], Wang [REDACTED], Li [REDACTED], and Huang [REDACTED] worked at A Company, they were all members of the fingertip recognition project team and all had opportunities to access technical information related to the secret points at issue.

The evidence submitted by the five alleged infringers was insufficient to prove that they had not committed acts of infringement against A Company's technical secrets at issue. First, according to B Company, the initial source of the accused infringing technology was mainly relevant open-source code collected by B Company from various open-source channels. However, how to integrate such open-source code obtained from different channels to form an initial model, and how to further form a well-performing model through extensive subsequent training, constitute both key stages and core difficulties in research and development. From obtaining open-source code related to fingernail recognition technology from different open-source channels to launching a product that can be commercially used in the market, a large amount of adaptation, integration, and refinement work undoubtedly still needs to be completed. Taking the "finger fingertip recognition technology" claimed by A Company as a trade secret in this case as an example, the research and development process from conceptual discussion to the final launch of the finished product took more than one year and five months. By contrast, B Company took less than two months from its establishment to providing its business partner with technology capable of realizing "finger positioning and recognition" functions. It is contrary to the rules of common experience for a product with artificial intelligence visual recognition and positioning functions to be launched from scratch within such a short research and development cycle, or at such a rapid pace of research and development. Second, both the "finger fingertip recognition" technology claimed and protected by A Company as a technical secret and B Company's accused infringing technology inevitably involve such procedural actions as recognizing a finger, positioning the finger, recognizing the text pointed to by the finger, correctly understanding the text content, and making a response. The aforementioned series of "actions" is in fact completed with the participation of artificial intelligence. For an "artificial intelligence" product to possess capabilities similar to human visual perception, identification, understanding, and response, it undoubtedly requires high-frequency and high-intensity training of the "artificial intelligence" model at the preliminary stage, such as training on how to recognize fingers. Without the "feeding" of data, an artificial intelligence model cannot function effectively. The richer, larger in scale, and higher in quality the data "fed" to an artificial intelligence model are, the stronger its self-learning and generalization capabilities will be, and the better its demonstrated cognitive level and processing capability will be. In the absence of sufficient "feeding" data to assist the artificial intelligence model in reinforcement learning and training, it is plainly unreasonable that B Company, relying solely on open-source code obtained from different open-source channels, could, within such a short period of time from May 2019 to July 2019, assist its business partner You [REDACTED] in launching a new product that differed from the original functions of You [REDACTED]'s product before the cooperation between the parties, had the new function of "finger recognition and point-reading," and was claimed to use "fingernail" technology "completely different" from A Company's "fingertip recognition and positioning" technology. Third, generally speaking, the knowledge and capabilities of an artificial intelligence model derive from the data used to train it. The training process of the model is the process of enabling it to learn to map input data to specific labels, and the capabilities of the model are strictly limited by the task boundaries defined by the training data. An artificial intelligence model does not acquire, out of thin air, the ability to recognize categories on which it has not been trained. Fundamentally, the quality of a model's output depends on the type, scale, and quality of the training data input into the model. According to the comparative demonstration video in which A Company's product and the accused infringing product faced the same tester and the same testing material at the same time, together with the facts further ascertained in the second instance, under the testing scenario where the fingernail was concealed, the accused infringing product supplied with technology by B Company could still successfully recognize, trigger positioning, and correctly spell and output the relevant words. This fact precisely shows that B Company's assertion that the accused infringing technology adopted "fingernail recognition" technology lacks persuasiveness. Finally, the "[REDACTED] English Reading Companion" is a column-shaped product equipped with a camera device, and objectively there is a possibility that users may move and place the column-shaped product at will during use. Therefore, when the product is used, its background system needs to perform angle adjustment work to collect samples. By contrast, the accused infringing product is a tablet-shaped product, and its usage characteristics basically do not require the product to adjust angles in the background. B Company also stated that its technology did not have an angle adjustment function. However, according to the product testing results, the accused infringing product could also recognize words when the angle was adjusted, namely, in a scenario involving an abnormal usage habit where an angle existed between the tablet and the user. Based on the above analysis, A Company's claim that B Company and the other four alleged infringers committed acts of misappropriating the trade secrets at issue should be upheld. Accordingly, the original judgment was revoked, and the judgment was amended to order the five alleged infringers to cease disclosing, using, and permitting others to use the trade secrets at issue, destroy the carriers bearing the trade secrets at issue, and bear joint and several liability for compensation in the amount of RMB 500,000.

This case provides relatively in-depth analysis and reasoning on how to apply Article 32, Paragraph 2 of the Anti-Unfair Competition Law, as amended in 2019, and how to allocate the burden of proof in handling disputes over misappropriation of technical secrets. It is of certain reference significance for the trial of cases concerning artificial intelligence and algorithm-related technical secrets.