GoKawiil
Tech News
Nvidia announced the Open Agent Safety Platform, which combines an open-source runtime called OpenShell with a hardware-based monitoring layer called Sentry that runs on its BlueField-4 data processing units. The system is designed to define, monitor and restrict what AI agents can do from testing through deployment, addressing incidents where agents bypassed security controls or acted outside their intended scope.
Nvidia has introduced a new security platform combining two tools, OpenShell and Sentry, designed to limit what AI agents can access, track their behavior, and terminate them if they attempt to exceed set boundaries. The announcement comes amid a rise in reported security incidents involving AI models behaving unpredictably or exploiting system permissions.
Nvidia announced its Open Agent Safety Platform, which it says can quarantine AI agents that breach their boundaries within milliseconds. The system pairs Nvidia's open-source OpenShell software, running on its Vera AI CPU, with a separate Sentry chip that continuously monitors agents and enforces access restrictions before and during tasks. Anthropic, Microsoft and SpaceX are among the companies backing the platform.
Nvidia has moved its OpenShell agentic AI sandbox, first unveiled at GTC in March, into general availability for all users. The tool isolates AI agents' activity at the operating system kernel level while they perform tasks, and Nvidia says it has security collaborations with companies including Anthropic, Cisco, CoreWeave, Dell, Microsoft, Palantir, Salesforce and SAP.
Nvidia has introduced the Open Agent Safety Platform, pairing its open-source OpenShell tool with a hardware watchdog called Sentry to police AI agent behavior. OpenShell enforces policies and tracks agent actions with minimal overhead on Nvidia's Vera CPUs, while Sentry runs on separate BlueField-4 data processing units to isolate controls from the agent's own system and can reportedly quarantine misbehaving agents within milliseconds.
OpenShell published research on using formal methods, including the Z3 solver, to verify that permission changes proposed by autonomous AI agents remain within the bounds originally approved by a human operator. The team argues that as organizations scale from a handful of coding agents to hundreds or thousands running long, open-ended tasks, manual permission review becomes impossible to sustain. Their approach aims to mathematically prove that scoped agent policies never exceed the intent of the overall system's charter.