Google opens HEIR for private AI inference

Google has open-sourced HEIR, a compiler that converts pre-trained models to run inference on encrypted inputs, and demonstrated it across 4 applications: recommendations, fraud detection, network intrusion, and hotword detection. Google’s announcement says the tool has already supported four peer-reviewed publications and collaborations with universities and hardware accelerators; the HEIR project is aimed at making homomorphic-encryption inference approachable beyond specialist cryptographers. Teams working with sensitive health data in AI products should file this as a compiler and research milestone, not yet a drop-in replacement for ordinary cloud inference: Google says the demos report single-threaded CPU latency and still describes accelerator benefits as forthcoming.