NVIDIA launches platform to keep AI agents inside security boundaries

NVIDIA introduced Open Agent Safety Platform, an open software platform and reference design for securing AI agents from testing through deployment. The launch is aimed at a problem that is becoming more practical than theoretical: agents are starting to operate across tools, infrastructure and, in some cases, robotics systems where application-level safeguards may not be enough.

The platform has two main pieces. OpenShell is open source runtime software that traces agent actions and enforces policy boundaries while agents run on CPUs. NVIDIA says it is optimized for its Vera CPUs but can be extended to third-party compute platforms, including Arm and Intel.

Sentry is the hardware-side reference design. It runs on NVIDIA BlueField-4 DPUs as an out-of-band watchdog, monitoring agent behavior from a separate trust domain. NVIDIA says Sentry can quarantine agents in milliseconds if they try to move outside permitted boundaries.

The company is positioning the stack as infrastructure for enterprises that want stronger controls outside the model itself. Anthropic is listed as a collaborator for Claude Managed Agents, and NVIDIA also names Microsoft, Salesforce, SAP, Hugging Face, Scale AI, ServiceNow, Palantir, CrowdStrike, Cisco, JPMorganChase and others as participants or adopters.