DESILO and Cornami Announce Breakthrough in Encrypted AI Computation

The new method accelerates encrypted matrix multiplication, advancing practical fully homomorphic encryption (FHE) for AI

DESILO, a Privacy Enhancing Technology (PET) company, and Cornami, a leader in scalable compute acceleration, announced new research that significantly improves the performance of encrypted AI computation using fully homomorphic encryption (FHE).

The research paper — co-authored by Craig Gentry, widely recognized as the father of FHE and a Gödel P**** laureate, and Yongwoo Lee, Head of Cryptography at DESILO — introduces a new method designed for efficient encrypted matrix arithmetic. According to results reported in the paper, the approach delivers up to 80× faster encrypted matrix multiplication compared to representative state-of-the-art baselines, marking a meaningful step toward in practical deployment of privacy-preserving AI.

Matrix multiplication is the computational backbone of modern machine learning models. The new method is designed to provide an optimized pathway for encrypted matrix multiplication across diverse scales and real-world workloads, narrowing the long-standing gap between theoretical cryptographic schemes and operational AI systems.

Also Read: CIO Influence Interview with Duncan Greatwood, CEO at Xage Security

“This research shows that privacy-preserving computation can be both efficient and practical,” said Seungmyung Lee, CEO of DESILO. “It forms a critical foundation for our encrypted AI stack, which enables organizations to analyze sensitive data without exposing it.”

The work reflects the strategic collaboration between DESILO and Cornami, combining DESILO’s advances in FHE-based computation with Cornami’s high-performance compute architecture. The joint effort focuses on making FHE usable in enterprise and AI environments where both data confidentiality and computational efficiency are essential.

“For decades, Fully Homomorphic Encryption (FHE) has been the gold standard for data privacy, but its computational cost has made real-world use impractical,” said Dr. Craig Gentry, Chief Scientist of Algorithms at Cornami.

Our latest research focuses on accelerating encrypted matrix multiplication, a fundamental operation representing over 90% of AI workloads. Combined with Cornami’s scalable Compute Fabric, it delivers orders-of-magnitude faster encrypted processing.”

“This makes privacy-preserving AI practical,” added Gentry. “Efficient, secure matrix multiplication enables Large Language Models to run inference on encrypted data with near-plaintext performance, strengthening compliance, sovereignty, and post-quantum security.”