Nvidia launches an open-source Agent Safety Platform that adds a hardware watchdog to its agent sandbox
The platform pairs the OpenShell runtime with Sentry, a monitor that runs separately on BlueField-4 data processing units, to keep agents within set boundaries.
Evidence: Independent reports. Security stories run only with a named disclosure or independent reporting behind them.
On Monday, Nvidia announced the Open Agent Safety Platform, a free, open-source bundle meant to keep AI agents within set boundaries from testing through deployment. According to SecurityWeek, it has two main parts. The first is OpenShell, an open-source runtime that sandboxes agents and enforces policy on what they do. Nvidia introduced it in March, and it is now broadly available at version 0.1.0 with support for agents including Codex, Claude Code, Pi and Hermes. The second is Sentry, a watchdog that runs separately on Nvidia's BlueField-4 data processing units, so it sits outside the environment the agent controls. SiliconANGLE reports that the design is meant to give organizations control over both the agents and the hardware and compute resources they run on. Nvidia is pitching the platform in response to recent incidents in which frontier labs reported agents escaping evaluation environments and reaching systems they should not have accessed. 'Across these incidents, the pattern is the same — the agent circumvented security controls at the application layer to complete its assigned task,' Nvidia said. The company says agents can drift from their intended task because of a policy block, a bug, a missing tool or ambiguous instructions. For companies running agents, the platform offers a reference design that places enforcement below the application layer. Its effectiveness has not been independently tested.