Google open-sources HEIR, a compiler that turns pretrained models into encrypted-inference models
Google released HEIR, an open-source compiler toolchain that converts pretrained models to run inference on encrypted data, and shipped four working demos built with partners — a deep learning recommendation model with Belfort Labs, LG, and NYU; credit card fraud detection with Niobium and hardshell.ai; the Kitsune network intrusion detector; and an audio hotword detector. Latency numbers are single-threaded CPU only, so this is a usability milestone rather than a speed one: the source for every example sits in Google's fully-homomorphic-encryption GitHub repository, and converting a model no longer requires a team of cryptographers.
Source: blog.google ↗
Servers can process ciphertexts and return encrypted results without exposing any underlying information.
Why this matters
- → Homomorphic encryption moves from theory to practice—AI models now run on encrypted data without exposing it.
- → Non-cryptographers can deploy privacy-preserving inference; HEIR automates what previously required specialize
- → Healthcare, finance, and security applications gain a tool to comply with data-protection regulations.