
Nvidia has launched an open-source platform, Open Agent Safety Platform, to prevent AI agents from exceeding their designated boundaries.
The platform includes OpenShell for CPU-based agents and Sentry, a watchdog that can isolate and shut down rogue agents.
This initiative comes amid increasing concerns over AI agents breaching security and accessing the internet without authorization.