Breakthrough in 'Private AI' Encryption

Fully Homomorphic Encryption (FHE) has been the holy grail of cryptography, a concept first proposed in 1978 and finally made viable in 2009 by Craig Gentry. His initial breakthrough, a first-generation FHE scheme, proved it was possible to compute on encrypted data, sparking decades of research to make it practical. The primary obstacle to widespread FHE adoption has been its immense computational overhead, making it orders of magnitude slower than operating on unencrypted data. This performance lag has been a major barrier for data-intensive fields like AI, where complex models require trillions of operations. Previous generations of FHE schemes, while improving efficiency, still struggled with the performance of specific operations crucial for AI. The new 5th-generation Gentry-Lee (GL) scheme, introduced at the FHE.org 2026 Conference, specifically targets the bottleneck of matrix multiplication, a foundational mathematical operation for training and running AI models like LLMs. The GL scheme, co-authored by Gentry and DESILO's Chief Scientist Yongwoo Lee, fundamentally restructures how homomorphic operations handle these matrix calculations. This architectural innovation is designed to make "Private AI" a functional reality, where AI models can process sensitive, encrypted data without ever needing to decrypt it. This breakthrough has significant implications for highly regulated industries. In healthcare, it could enable AI-powered analysis of sensitive patient data for medical research without compromising patient privacy. For financial services, it allows for fraud detection and risk modeling on encrypted financial data, maintaining confidentiality and compliance. The move towards practical FHE is also critical in the context of the M&A landscape, where technology and data assets are central to deal value. The global M&A market saw a rebound in 2025, with a notable increase in large-scale transactions, particularly in the technology and financial services sectors. As AI continues to drive strategic decision-making, the ability to securely leverage sensitive datasets for competitive advantage will be paramount. Technologies like the GL FHE scheme could become a key enabler for unlocking value in future M&A, allowing for deeper due diligence on data-centric companies without exposing proprietary information.