#South Korea #Seoul #Private AI #Desilo #Fully Homomorphic Encryption

Introduction

In a significant development for privacy-enhancing technologies, DESILO has unveiled the Gentry–Lee (GL) scheme at the FHE.org 2026 Conference in Taipei. This fifth-generation Fully Homomorphic Encryption (FHE) scheme, co-authored by distinguished scientists Craig Gentry and Yongwoo Lee, aims to optimize the crucial operation of matrix multiplication, particularly beneficial for artificial intelligence (AI) systems.

The Need for Private AI

As artificial intelligence solutions permeate various industries, data privacy has emerged as a pressing concern. Private AI represents a paradigm shift that allows AI systems to process encrypted data directly. This means sensitive information can remain protected while still being utilized to enhance services and products. Industries with strict regulatory requirements, such as healthcare and finance, greatly stand to benefit from these advancements, as they handle sensitive personal data.

Understanding Fully Homomorphic Encryption

Fully Homomorphic Encryption is a revolutionary technique that permits computations to be performed on encrypted data without decrypting it first. Although earlier FHE schemes were instrumental in promoting encryption use, they still faced challenges related to computational efficiency and overhead, especially for modern AI applications. The GL scheme promises to address these issues, making it a significant advancement in the field.

Enhancements Introduced by the GL Scheme

The GL scheme is designed to significantly improve matrix multiplication efficiency, an operation heavily utilized in today's AI systems, including Large Language Models (LLMs). Craig Gentry states that initial efforts in FHE focused primarily on mathematical proofs of feasibility, whereas the GL scheme redefines how homomorphic operations conduct matrix multiplication. This innovation is pivotal in bringing encrypted computations closer to practical applications in advanced AI systems.

Impact on Matrix Multiplication

Matrix multiplication serves as the backbone of many AI algorithms. By enhancing this fundamental operation, the GL scheme facilitates more efficient processing of encrypted data, thus supporting Private AI. Yongwoo Lee emphasizes that this advancement significantly enhances how such operations are executed under homomorphic encryption, pushing practical applications of Private AI nearer to reality.

Research and Implementation

A comprehensive examination of the Gentry-Lee scheme is available in the technical paper published through the IACR ePrint archive. Researchers and developers interested in understanding the intricacies of this new scheme can access the full details online. This collaborative effort marks a crucial stage in the journey of private AI, ultimately aiming to balance the benefits of technological advancement with the imperative of data privacy.

Conclusion

As organizations increasingly adopt AI technologies in their operations, the demand for secure methods to handle sensitive data will likely escalate. The Gentry-Lee scheme presents a promising solution tailored to meet these needs, demonstrating DESILO's commitment to pioneering innovations in privacy-aware technologies. This milestone not only sets the stage for developments in encryption-based AI methods but also highlights the importance of protecting user data in an era increasingly driven by information security concerns.