NVIDIA launches open AI agent safety platform
- NVIDIA introduced its Open Agent Safety Platform, combining runtime software and a reference design to help manage AI agent behavior.
- The open-source OpenShell runtime tracks agent actions and applies policies, while Sentry can isolate agents that cross software boundaries.
- NVIDIA said companies including Salesforce and SpaceXAI are using components of the platform for AI agent applications.
NVIDIA (NASDAQ:NVDA) introduced its Open Agent Safety Platform, an open-source framework designed to provide controls for AI agents through runtime monitoring and policy enforcement tools.
The platform includes OpenShell, a runtime system that traces agent actions and applies security policies, and Sentry, a reference design that uses BlueField-4 processors to monitor and quarantine agents that cross software boundaries.
NVIDIA said OpenShell runs on NVIDIA Vera CPUs and can be extended to Arm and Intel compute platforms, while Sentry is designed to isolate certain agent activities within milliseconds when policy boundaries are exceeded.
NVIDIA develops accelerated computing hardware and software used in artificial intelligence, data centers, graphics, and other computing applications.
The company said Salesforce has integrated OpenShell with Slack, while SpaceXAI is using the platform for Cursor coding agents and Grok models, with platform software available through NVIDIA developer resources and GitHub.
