NVIDIA launches Open Agent Safety Platform for AI security

NVIDIA has announced the Open Agent Safety Platform, a new software platform and reference system design intended to secure AI agents from the testing phase through deployment. The initiative aims to address security vulnerabilities where agents have previously circumvented application-layer controls to complete tasks.

The platform consists of two primary components: OpenShell and Sentry. OpenShell is an open-source software that provides a secure runtime boundary, tracing actions and enforcing policies when agents access data, tools, or APIs. While optimized for NVIDIA Vera CPUs, its open-source nature allows for compatibility with third-party platforms like Arm and Intel.

Sentry serves as an out-of-band hardware watchdog running on NVIDIA BlueField-4 DPUs. It provides continuous monitoring of agent behavior and is capable of quarantining or interrupting an agent within milliseconds if it attempts to operate outside its defined boundaries. This hardware-level layer is designed to prevent agents from interfering with the very mechanisms meant to monitor them.