At the 36Kr WAVES 2026 conference, Yu Gongshan, co-founder of Chenyi Technology, delivered a keynote on how full homomorphic encryption (FHE) can address data security challenges in the AI era. He warned that current computer security systems are vulnerable to advanced AI models like those from Anthropic, and proposed a new cryptographic approach to enable computation on encrypted data without decryption.
The Data Security Dilemma in AI
Yu highlighted a real-world problem: banks need to use their data to train AI models for risk management, but fear that models might memorize or leak sensitive information. This creates a paradox where enterprises want to leverage new computing power for AI but are constrained by security concerns. He compared data to nuclear energy—immensely valuable but dangerous if exposed. The solution, he argued, is to build a "data containment chamber" that allows computation without revealing the underlying data, much like a nuclear reactor that converts energy without releasing radiation.
Full Homomorphic Encryption as the Foundation
Traditional security requires decrypting data before use, which risks leakage. Full homomorphic encryption (FHE), a concept dating back to the 1970s, enables computation on encrypted data. However, early FHE algorithms suffered from massive performance overhead, hindering commercial adoption. Chenyi Technology has been researching FHE since the 1990s and claims to have developed a practical solution: a multi-modal database that integrates vectors, graphs, time-series, and relational data, all while operating on encrypted data. Yu gave an example of identifying conference attendees by combining encrypted video feeds, social security records, and internet posts without ever decrypting the data.
Key Advantages of Chenyi's Solution
Yu outlined five features of their product. First, it is 100% self-developed with zero open-source code, from mathematical theory to the underlying compiler. Second, it achieves 37% higher performance than traditional plaintext databases while reducing hardware costs by 62%. Third, it is natively Chinese-language, allowing users to interact with the database using Chinese commands, lowering the barrier for non-technical users. Fourth, it natively supports AI multi-modal processing. Fifth, it uses lattice-based cryptography to resist future quantum computer attacks, ensuring security for at least 30 years.
Applications and Vision
The technology is already used in finance, but Yu envisions broader applications. He proposed creating secure data chambers for every city, enabling "data available but invisible" for urban data sharing. This would allow cities to interconnect their data safely, unlocking value in healthcare, supply chains, and insurance. For example, medical insurance data could be combined with commercial insurance data in an encrypted environment to enable real-time settlement at hospitals. Yu called for industry-wide adoption of FHE to make data as easy to use as electricity, driving innovation while maintaining security.