DESILO's THOR Selected as Reference for Encrypted AI in Global FHE Benchmarking
DESILO Inc. announced Tuesday that its THOR system has been chosen as the reference implementation for a global benchmarking suite for encrypted AI. THOR is the first framework capable of running the entire LLM inference process directly on encrypted data.
DESILO Inc. announced Tuesday its THOR system has been selected as the reference implementation for a global benchmarking suite focused on Fully Homomorphic Encryption (FHE). The initiative aims to establish a universal benchmark for private AI and encrypted Large Language Model (LLM) inference.
THOR is described as the world's first framework that allows the entire LLM inference process to operate directly on encrypted data. This capability is intended to eliminate the need for intermediate decryption and re-encryption steps, thereby enhancing data security and privacy for AI applications, especially those handling sensitive information.
The selection for the benchmarking suite was made in late July. This move is expected to contribute to the advancement and adoption of decentralized and privacy-preserving AI solutions within the industry.