In a trade secret infringement ruling, the Supreme People's Court clarified the burden of proof in such cases: when a rights holder provides preliminary evidence reasonably indicating infringement, and also demonstrates that the alleged infringer had access to the trade secret and that the information used is substantially identical to it, the burden shifts to the alleged infringer to prove that no infringement occurred. If the alleged infringer asserts independent development of the accused technology, the court must conduct a thorough and objective examination of the supporting evidence, applying logical reasoning and everyday experience. This examination should take into account the specific performance characteristics of the accused product and compare them with those of the rights holder’s product embodying the trade secret.
This case concerns a dispute over trade secret infringement in the visual recognition sector, involving artificial intelligence and algorithmic technology. The facts are summarized below:
Company A focused on developing a product called the "English Reading Companion," which offered fingertip recognition and point-to-read translation that enables users to instantly recognize and translate printed text by simply pointing at it with their finger. Company A filed a lawsuit with the court of first instance, seeking an order requiring Company B, along with individuals W, X, Y, and Z, to immediately cease the infringement and to pay joint compensation for economic losses.
According to Company A, it had established the relevant trade secrets - comprising an algorithm based on fingertip recognition code and an associated image database - by April 2019. W, a former shareholder and Chief Technology Officer of Company A, along with former employees X, Y, and Z, all worked on the "fingertip recognition" project and had access to the technical information at issue. W left Company A in March 2019 and founded Company B two months later. X, Y, and Z subsequently left Company A and joined Company B as shareholders or key technical staff.
Beginning in July 2019, a series of products featuring fingertip recognition and point-to-read functionality, powered by Company B's technology, were introduced to the market. Company A claimed that W, X, Y, and Z had, without authorization, disclosed its trade secrets to Company B and permitted their use. Company B then leveraged the secrets to provide technical support to other companies, which in turn launched products with similar fingertip recognition and translation features.
The court of first instance dismissed Company A’s claims, ruling that the evidence presented failed to establish that the technical information used in the accused products, which incorporated Company B’s technology, was substantially identical to the technical information claimed by Company A as its trade secret. The court further observed that the technical information used in the accused products differed materially from the trade secret information claimed by Company A. As a result, the evidence on record could not reasonably indicate that Company A’s claimed technical information had been infringed.
Dissatisfied with the judgment, Company A appealed to the Supreme People's Court.
On appeal, the Supreme People’s Court overturned the first instance judgment and ruled that the five alleged infringers must cease disclosing, using, or authorizing others to use the trade secrets, destroy all carriers containing such secrets, and bear joint and several liability for compensation. The appellate court found that Company A’s evidence reasonably established infringement by the five parties.
First, on its corporate homepage dated October 14, 2019, Company B expressly featured the “English Reading Companion” developed by Company A as a flagship product of its “Desktop Interactive Technology Platform,” highlighting that the product integrated AI-based fingertip-positioning technology capable of real-time finger recognition and accurate word translation.
Second, in an earlier action brought by Company A against Company B on substantially identical facts (later withdrawn), a defense initially filed by W showed that, in his own contemporaneous understanding, Company B’s product used the same “finger recognition and tracking” technology as Company A’s, with both products exhibiting the ability to quickly recognize English words based on finger position.
Third, side-by-side comparison demonstrations revealed nearly identical recognition, output, and pronunciation performance between the accused product and Company A’s product, regardless of whether finger interaction was used in the test.
Fourth, the interval between the founding date of Company B’ (May 21, 2019) and the date its partner’s product first demonstrated finger-positioning and recognition capability “from scratch” (July 4, 2019) was less than two months.
Finally, W, X, Y, and Z had all served on the fingertip-recognition project team during their tenure at Company A, thereby having access to the technical information that constituted the trade secret.
The evidence submitted by the five accused infringers was insufficient to prove that they did not infringe upon Company A's trade secrets.
First, Company B contended that the accused technology was derived primarily from open-source code obtained through various channels. However, integrating code from disparate sources into a functional initial model, then refining it through extensive training to achieve commercial readiness, constitutes a critical and highly complex phase of development. Transforming open-source “fingernail-recognition” code into a market-ready product inevitably requires substantial adaptation, integration, and polishing. For instance, Company A’s “fingertip recognition technology”, asserted here as a trade secret, underwent over seventeen months of development from concept to product launch. By contrast, Company B supplied its partner with operational “finger positioning and recognition” technology less than two months after its founding. Developing an AI-based visual recognition and positioning product from scratch in such a short period runs counter to ordinary experience.
Second, both Company A’s protected “fingertip recognition technology” and Company B’s accused technology necessarily involve steps such as finger detection, positioning, recognition of pointed-to text, comprehension of content, and generating appropriate responses, which are all driven by AI. Equipping an AI product with human-like visual perception, recognition, comprehension, and response abilities demands intensive, high-frequency prior training, including training for finger recognition. Without “feeding” data, an AI model cannot operate effectively; the richer, larger, and higher-quality the training data, the stronger the model’s self-learning, generalization, and performance become. In the absence of sufficient training data to enable reinforcement learning, it defies logic that Company B could, relying solely on open-source code, help its partner (Company C) launch, within a few months (May–July 2019), a new “finger recognition and point-to-read” product that differed functionally from Company C’s earlier offerings, while simultaneously claiming the technology was “completely different” from Company A’s “fingertip recognition and positioning” and instead calling it “fingernail” technology.
Third, an AI model’s knowledge and capabilities fundamentally come from its training data. The training process teaches the model to map inputs to specific labels, and the model’s abilities remain strictly bounded by the tasks defined in the training data. An AI model cannot spontaneously recognize categories it has never been trained on. Ultimately, output quality depends on the type, scale, and quality of the training data. Comparative demonstration videos - in which both Company A’s product and the accused product were tested by the same person using the same materials - showed that the accused product, powered by Company B’s technology, could still accurately recognize, locate, pronounce, and output words even when the fingernail was concealed. This observation directly undermines Company B’s assertion that its technology relies on “fingernail recognition.”
Finally, the “English Reading Companion” is a cylindrical device with a camera, meaning users may freely reposition it during operation. Its backend therefore requires angle-adjustment functionality to capture usable samples. The accused product, however, is a tablet-style device whose typical use does not require such adjustment, and Company B stated its technology lacks angle-adjustment capability. Yet tests revealed that the accused product could still recognize words even when tilted (i.e., under non-standard angled conditions).
Based on the foregoing analysis, the court concluded that Company A’s claim of trade-secret infringement by Company B and the four individual defendants was well founded.
Through this ruling, the Supreme People's Court has clarified the allocation of burden of proof in trade secret disputes involving AI technologies: The rights holder needs only to establish a preliminary evidentiary chain, supported by R&D records, data scale, and functional comparisons, to trigger the burden-shifting mechanism under Article 32 of the Anti-Unfair Competition Law. If the alleged infringer raises an “independent R&D” defense, it must produce code, data, and development‑history records that are both contemporaneous with and traceable to the challenged activities. A mere assertion of reliance on open‑source code will generally not suffice.
This judgement decision underscores that in the rapidly evolving, data‑intensive domain of AI, a well‑maintained, real-time, and fully traceable self‑documentation system is the most dependable means for a company to protect its rights. Such a system also forms the critical factual basis on which courts assess whether the accused technology is substantially identical to the claimed trade secret.
(2023) Zui Gao Fa Zhi Min Zhong No. 1503
If you have any question about the protection of intellectual property rights, please feel free to send us emails. For patent-related matters, please send to info@afdip.com. For trademark/litigation/legal matters, please send to info@bhtdlaw.com.
