Nvidia has launched Open Agent Safety, an open-source reference platform designed to monitor and regulate the behavior of AI agents. The system acts as a supervisor, preventing agents from performing unauthorized actions or accessing restricted data within a corporate environment.
The platform features Nvidia OpenShell, an open-source runtime environment that manages agent permissions and interactions. It ensures that agents remain within defined operational boundaries, even if they exhibit unpredictable behavior, thereby containing potential errors.
For organizations requiring advanced security, the company introduced Nvidia Sentry. This tool provides out-of-band, chip-level telemetry to enforce security policies, enabling the isolation of AI agents within milliseconds.
While the new software suite is compatible with various hardware configurations, it is specifically optimized for systems utilizing Nvidia Vera processors and Nvidia BlueField DPU units. This development expands the company's hardware ecosystem.
