Nvidia's Agent Safety Platform Splits Enforcement into Hardware and Software

Nvidia's Open Agent Safety Platform is designed to enforce boundaries on AI agents through dedicated silicon. The platform has four components, including OpenShell, a secure runtime that can be extended to work with third-party compute platforms, and Sentry, an out-of-band watchdog that performs the actual enforcement.

The release highlights the difference between the open, portable component (OpenShell) and the enforcing component (Sentry). While OpenShell extends to Arm and Intel by Nvidia's description, Sentry's portability is not addressed in the release. This asymmetry is a design choice worth watching, as it could impact adoption across different ecosystems.

The platform also includes DOCA for inspection and enforcement logic, and the Vera CPU and BlueField-4 DPU are the underlying hardware. Nvidia announced the Open Secure AI Alliance, governed by the Linux Foundation, which counts over 120 leading organizations among its initiators.