NVIDIA announced its Open Agent Safety Platform on September 28, combining software controls with a hardware monitoring reference design for autonomous AI agents. The company says the approach places enforceable limits outside the agent itself.
OpenShell and Sentry serve different roles
OpenShell provides the runtime boundary: it traces actions and enforces policies as agents execute tasks. NVIDIA says the open-source software is broadly available and can be extended to third-party computing platforms, including Arm and Intel.
Sentry is the hardware component in the reference design. It runs on BlueField-4 data processing units and monitors activity separately from the agent’s host environment. NVIDIA says it can quarantine an agent that crosses its permitted boundaries within milliseconds; that is a vendor claim, not an independent guarantee.
Partners add their own controls
Check Point separately reported a beta integration connecting its semantic monitoring to OpenShell. Its monitor considers an agent’s task and action history, while OpenShell supplies an enforcement point before actions reach the host.
The distinction matters: infrastructure controls determine what an agent can access, while contextual monitoring assesses whether a permitted action still fits the assigned work. The launch brings these approaches together, but does not establish that every deployment or agent behaviour is safe.
Sources: NVIDIA announcement, republished by MarketScreener; Check Point integration announcement.
Feature image: AI-generated editorial illustration for Guava Insights; conceptual imagery, not a product or location photograph.
