Using a new Rust-based operating system, Google aims to secure ambient machine learning on embedded hardware.
Researchers at Google recently announced a mathematically-secure platform, KataOS, optimized for embedded ML applications. The Alphabet giant has shared some early details on this project (which is still under development) and is inviting others to collaborate on its open-source platform.
Google’s homegrown operating system KataOS is part of a larger project, Sparrow, that leverages RISC-V and Google’s hardware root-of-trust OpenTitan. The project aims to design a secure, low-power embedded platform for “ambient ML applications.” KataOS runs on top of seL4, one of the world’s fastest operating system kernels built for security, and is written almost entirely in the Rust programming language.