Google Launches HEIR Compiler for Private AI Inference

What happened

Google introduced HEIR, an open-source compiler in its Private Computing Toolkit, to enable cryptographically-secure private AI inference. It converts pre-trained models to run on encrypted data, allowing applications like fraud detection and recommendations without exposing user data. Partners include Belfort, Niobium, and academic institutions like Tsinghua University.

Why it matters

HEIR addresses the privacy-security trade-off in AI by allowing computations on encrypted data without decryption. This enables critical sectors like healthcare and finance to use AI features while complying with strict regulations, and allows cloud providers to offer services like spam detection without accessing raw data.