NEWS

Co-authored with FHE inventor Craig Gentry; simultaneous acceptance of both scheme and bootstrapping papers marks exceptionally rare academic feat Hundreds-fold speedup in matrix operations accelerates path to commercial Private AI SEOUL, South Korea , May 14, 2026 /PRNewswire/ -- DESILO, a specialist in Fully Homomorphic Encryption (FHE) and Privacy-Enhancing Technology (PET), announced today that two papers on its 5th-generation homomorphic encryption scheme 'GL,' led by DESILO researchers, have been simultaneously accepted at

DESILO, a pioneering deep-tech company specializing in privacy-enhancing technologies, has announced the release of the world's first Fully Homomorphic Encryption (FHE) library to seamlessly integrate the 5th-generation 'GL Scheme (Gentry-Lee Scheme)'. This breakthrough marks a monumental step forward in making 'Private AI'—the ability to train and run advanced AI models directly on encrypted data—a practical reality.

Debuting at the FHE.org 2026 Conference, the Gentry–Lee (GL) scheme introduces a breakthrough in matrix multiplication performance for Fully Homomorphic Encryption. SEOUL, South Korea and TAIPEI, March 7, 2026 /PRNewswire/ -- DESILO, a deep-tech company specializing in privacy-enhancing technologies, today unveiled the Gentry–Lee (GL) scheme, a major advancement positioned as the 5th generation of Fully Homomorphic Encryption (FHE). The scheme is being formally presented at the FHE.org 2026 Conference in Taipei. The framework is co-authored by Yongwoo Lee (Chief Scientist, DESILO) and Craig Gentry (Chief Scientist, Cornami) — the inventor of FHE in 2009 and recipient of the 2022 Gödel Prize. As organizations increasingly adopt AI while handling sensitive data, a new paradigm known as Private AI is emerging. Private AI enables AI systems to operate directly on encrypted data, allowing enterprises and organizations to use AI without exposing sensitive prompts, inputs, or model outputs to the AI provider. This capability is particularly important for highly regulated industries, enterprise environments with strict data protection and sovereignty requirements, and organizations working with highly sensitive or valuable data. Fully Homomorphic Encryption (FHE) enables this approach by allowing computations to be performed directly on encrypted data without decrypting it. While earlier FHE schemes made encrypted computation increasingly practical, their computational overhead has remained a major challenge for modern AI workloads. The GL scheme introduces a new architecture designed to significantly improve the efficiency of matrix multiplication, one of the core operations underlying modern deep learning systems. Because Large Language Models (LLMs) rely heavily on repeated matrix multiplication, improving this operation is critical for enabling AI computation under encryption. "Back in the early days of FHE, the primary goal was proving mathematical feasibility," said Craig Gentry, Chief Scientist of Cornami. "With the GL scheme, we are fundamentally restructuring how homomorphic operations handle matrix multiplication. Optimizing these core operations brings encrypted computation much closer to supporting modern AI architectures." "Matrix multiplication is the dominant workload in modern AI systems," said Yongwoo Lee, Chief Scientist of DESILO. "With the GL scheme, we introduce a new framework designed to significantly improve how these operations are performed under homomorphic encryption, bringing practical Private AI closer to reality." The full research paper describing the GL scheme is available via the IACR ePrint archive. Technical paper: https://eprint.iacr.org/2025/1935 FHE.org 2026 Conference: https://fhe.org/conferences/conference-2026/

SEOUL, South Korea, Nov. 12, 2025 /PRNewswire/ -- DESILO, a Privacy Enhancing Technology (PET) company, and Cornami, a leader in scalable compute acceleration, today announced new research that significantly improves the performance of encrypted AI computation using fully homomorphic encryption (FHE).