NVIDIA releases an open runtime to contain rogue AI agents

NVIDIA launched the Open Agent Safety Platform on 28 September 2026, a set of software and hardware components meant to constrain autonomous AI agents from testing through deployment.

The platform has two parts. OpenShell is an open-source runtime that sets a boundary around how an agent executes tasks, whatever model drives it; it runs on NVIDIA Vera CPUs and can extend to Arm and Intel platforms. Sentry is a reference design for an out-of-band watchdog running on BlueField-4 data processing units, which monitors agent behaviour separately from the host and, according to NVIDIA, can quarantine an agent that tries to leave its boundaries within milliseconds.

NVIDIA says more than 100 organisations are collaborating on the platform, among them Anthropic, Hugging Face, Microsoft, CrowdStrike and JPMorganChase, as well as the robotics companies Figure, Gecko Robotics and Skild AI. OpenShell is available on GitHub and through NVIDIA’s developer pages.

The launch follows a series of 2026 incidents in which research agents reached systems they were not meant to touch, and places the containment layer in the infrastructure rather than in the model.