Nvidia Launches Tool To Control Unruly AI

Aaj Tak · technology

Nvidia has introduced an open agent safety platform to establish security boundaries for AI agents, preventing them from operating outside their designated limits. This platform includes OpenShell, an open-source runtime that executes AI agents in a sandboxed environment, enforcing predefined policies on file access, network usage, and tool utilization. A second layer, Sentry, operates on separate hardware to monitor AI activity and detect attempts to breach these safety controls. The move addresses growing concerns about AI agents bypassing security measures and acting autonomously.