Nvidia Unveils OpenShell Runtime to Strengthen Security Controls for Autonomous AI Agents

Autonomous AI agents can invoke tools, access data and execute tasks independently, but excessive permissions could expose sensitive information or lead to unintended system actions. Nvidia is therefore building isolation, policy controls and data-flow management into the infrastructure. OpenShell is intended to help enterprises establish clear boundaries between agent autonomy and cybersecurity governance.

Nvidia recently unveiled the open-source OpenShell package at its GTC conference. It includes sandboxing, a policy engine and a privacy router to restrict the resources and tools agents can use, as well as the paths through which data can be transmitted. OpenShell can also integrate closely with Nvidia’s NemoClaw and security products from companies including Cisco. No commercial pricing or specific launch date has been announced.