DESILO and Cornami Unveil Encrypted AI Model at AI Infra Summit

DESILO and Cornami have launched a fully homomorphic encryption (FHE)-based large language model (LLM) at the AI Infra Summit 2025, announced in a press release. This new solution processes sensitive data while it remains encrypted, offering both speed and accuracy in real-world applications.

The collaboration addresses the long-standing tradeoff between data protection and AI performance, particularly in sectors like healthcare and finance. Traditional encryption methods often slow down computation, but the Cornami-DESILO model enables encrypted AI inference at speeds comparable to plaintext processing. Dr. Craig Gentry, Chief Scientist of Algorithms at Cornami, emphasized the role of plaintext ciphertext matrix multiplication (PCMM) in achieving this efficiency.

The encrypted AI model is particularly beneficial for industries requiring stringent data security, such as healthcare, finance, and government. It allows for the analysis of encrypted data without exposing sensitive information, aligning with compliance and data sovereignty requirements. This initiative also supports the upcoming launch of DESILO's HARVEST™ platform, which aims to enhance secure multi-party data collaboration globally.