Nvidia Unveils Open Security Platform for AI Agents

On September 28, Nvidia launched an open agent security platform designed to provide continuous monitoring, real-time policy enforcement, and security governance for AI agents through application, runtime, and infrastructure layers.

The platform includes two key components. OpenShell, an open-source tool, runs AI agents in a sandbox environment, converting user instructions into verifiable security policies and restricting agent access to files, networks, tools, processes, and credentials. Nvidia Sentry extends monitoring and security controls to BlueField data processing units, using Nvidia DOCA to correlate agent interactions, policy decisions, and tool/data access records to identify anomalous behavior and intervene when necessary.

Nvidia noted that AI agents may deviate from intended tasks due to factors such as tools, runtime, or ambiguous instructions, which calls for security controls independent of the agents themselves. The platform is optimized for Nvidia Vera CPU and BlueField DPU systems, while remaining compatible with other hardware platforms.