Nvidia Unveils Open Agent Safety Platform for Secure AI
Nvidia has extended its controls for AI agents by introducing the Open Agent Safety Platform. This platform combines Nvidia's OpenShell runtime with Sentry, a reference design for monitoring agents independently of host systems.
The OpenAgent Safety Platform is designed to pair runtime restrictions with an independent hardware monitoring layer. It allows organizations to restrict access to files, processes, networks, tools, services, and credentials.
Nvidia tested the platform in an adversarial experiment comprising ten 30-minute sessions and two 2-hour sessions. The results showed that an AI reviewer made 416 decisions and mistakenly approved one malformed proposal containing write authority.
Sentry is intended to add enforcement from a separate hardware layer, inspecting activity, verifying an agent's identity and delegated authority, enforcing access policies, and quarantining a workload within milliseconds.