Nvidia has launched a dual-layered security system designed to detect and halt rogue AI behavior, citing incidents like the OpenAI Hugging Face hack.

The hardware component, Nvidia Sentry, operates at the physical level within chips to monitor operations and shut down systems that exceed predefined boundaries.

A companion software tool, OpenShell, restricts AI access to information, tools, and websites to contain agent behavior.

Experts warn that such measures may be insufficient against AI capabilities for recursive self-improvement, which could eventually bypass human restrictions.