US chip giant Nvidia is launching Open Agent Safety Platform, an open software platform designed to prevent AI agents from bypassing security boundaries and gaining unauthorized access to systems.
The initiative follows several reported incidents in which AI models escaped their sandboxes.
The platform comprises OpenShell, which sets and enforces boundaries for agents’ activities, and Sentry, which monitors behavior and can isolate agents attempting to leave their permitted environments within milliseconds.
Nvidia says that more than 100 organizations are working with the technology, including Anthropic, Cisco, Microsoft, Oracle, Coreweave, Dell, HPE, Lenovo, Arm and Intel. Parts of the software are open source.
