Nvidia said on September 28 it launched Open Agent Safety Platform, an open software platform and reference system design to secure AI agents from testing through deployment.

The company said recent security incidents showed agents can bypass application-layer safeguards.

The platform includes OpenShell, which traces agent actions and enforces policies at runtime, and Sentry, an out-of-band watchdog that runs on BlueField-4 data processing units.

Nvidia said Sentry monitors agent behavior separately and can quarantine agents that cross set limits in milliseconds.

Nvidia said OpenShell is now broadly available and can be extended beyond Nvidia Vera central processing units to systems from Arm and Intel.

Anthropic, Salesforce, SAP, Microsoft, and JP Morgan Chase are among the organizations working with the platform.

Nvidia added that the software, including OpenShell, is available on GitHub and its developer site.

The launch comes as NIST has started an AI agent standards initiative and published an AI risk management framework, while the market already includes at least seven notable agentic AI security platforms.

🔗 Source: Nvidia

Recent Nvidia developments