Google is enhancing private AI capabilities with HEIR, an open-source compiler that enables cryptographically-secure AI inference. Homomorphic encryption allows computations on encrypted data, addressing privacy concerns in sectors like healthcare and finance where data sharing is restricted. While traditionally requiring specialized cryptographic expertise, HEIR aims to simplify this process, making it a "one-click solution" for non-experts to integrate encrypted inference into applications.

HEIR has gained traction within the homomorphic encryption community, with partnerships forming with hardware accelerator companies such as Belfort, Niobium, Cornami, and Optalysys. These collaborations are expected to yield significant latency improvements. The compiler also serves as a research platform, fostering collaborations with numerous universities and leading to several peer-reviewed publications.

To showcase HEIR's potential, Google has demonstrated four private inference applications: a recommendation model (with Belfort Labs, LG, and New York University), a credit card fraud detector (with Niobium and hardshell.ai), a threat intrusion detection system (with Niobium), and a hotword detector (with Belfort Labs). These examples highlight the practical application of homomorphic encryption, with latency figures provided for single-threaded CPU execution. Google plans to continue advancing homomorphic encryption to make it more accessible and efficient across industries.