Separate safety hardware that watches AI agents and can force-stop them when they break the rules.
It is like a lifeguard at the pool. It never gets in the water. It just watches, and pulls you out the second you cross the rope.
It runs on its own hardware. The agent cannot switch it off.
AI sandbox
Nvidia's OpenShell keeps agents in a sandbox. The Sentry watchdog watches from separate hardware outside it.
AI kill switch
When an agent crosses the line, the watchdog isolates and stops it within milliseconds.
Agent security
It moves safety checks into hardware, so even a hacked agent cannot get around them.