Nvidia has launched the Open Agent Security Platform, providing two-layer security protection for AI agents through open-source tools OpenShell and Nvidia Sentry. The system aims to enable safer testing of advanced AI systems by restricting access permissions and isolating anomalous behavior in real time. OpenShell runs on Nvidia's Vera CPU to restrict AI agent access permissions, while Nvidia Sentry operates on BlueField data processing units to monitor agent behavior. Sentry can detect anomalies and isolate agents in as little as a few milliseconds, offering rapid response capabilities for securing autonomous AI operations.