Google's HEIR Aims to Make Homomorphic-Encrypted Inference a One-Click Capability

Google has launched HEIR, an open-source compiler toolchain that enables AI models trained on unencrypted data to perform inference on homomorphically encrypted inputs, enhancing privacy and security. HEIR supports Python programming with encrypted data annotations and can compile PyTorch models for encrypted execution, though current performance overhead remains a challenge.