Real-Time FHE Applications and the Era of Encrypted AI

No ratings

Presented at Real World Crypto 2026 by

Jung Cheon

Since its first implementation in 2011, Fully Homomorphic Encryption (FHE) has experienced rapid growth, with processing speeds improving sevenfold annually even without customized hardware. Consequently, many use cases have now reached the stage of real-time feasibility. In this presentation, we introduce the latest performance benchmarks for FHE operations and bootstrapping on CPU/GPU platforms and their applications including encrypted statistics and machine learning. In particular, we highlight recently proposed homomorphic matrix operations and their real-time applications in fields such as encrypted RAG and search. Finally, we examine the current implementation trends and practical applications of encrypted Large Language Models (LLMs).