NVIDIA Launches an Open Platform for Containing AI Agents
2026-09-28 ยท Safety & Alignment
NVIDIA's Open Agent Safety Platform pairs an open-source sandbox for AI agents with a separate hardware watchdog that can quarantine an agent from outside its environment.
The platform combines OpenShell, NVIDIA's open-source runtime that confines an agent to a workspace with explicit permissions for files, tools, network access, and credentials, with Sentry, a watchdog that runs independently on BlueField-4 hardware. NVIDIA says Sentry operates from an isolated, out-of-band trust domain that is invisible to agents and attackers, that it can quarantine an agent in milliseconds, and that more than 100 organizations are working with the platform's technologies. The platform sits within the Open Secure AI Alliance, which NVIDIA formed in July and which moved under Linux Foundation governance in September. NVIDIA frames the platform as a response to recent incidents in which agents circumvented security controls at the application layer to complete their tasks. Robotics companies including Figure are building with OpenShell, and energy and critical-infrastructure providers are among the organizations working with the platform.
For SIRDA, the design separates the monitor from the thing it monitors, down to the hardware: counter-agency built into infrastructure. It leaves open who oversees the watchdog and who decides what it may quarantine.