NVIDIA Launches Platform to Enforce AI Agent Safety With Software and Hardware Controls

· Original article ↗

Summary

NVIDIA’s Open Agent Safety Platform combines OpenShell isolation and access policies with Sentry hardware monitoring to detect and stop AI agents that exceed their permissions.

Key points

  • OpenShell runs agents in isolated environments and controls access to files, networks, processes and other resources.
  • Sentry monitors agent activity from separate hardware using NVIDIA BlueField-4 DPUs and DOCA software.
  • Sentry can quarantine and stop agents that attempt to cross their software boundaries; it requires supported hardware.
  • NVIDIA says OpenShell can run on its Vera CPUs and can be extended to other platforms, including Arm and Intel.
  • Scale AI says it is using the reference design for enterprise and government applications.

Article Details

Event Type
NVIDIA launched the Open Agent Safety Platform, an open software platform and hardware-based reference design for AI agent safety.
Impact
The platform combines OpenShell runtime isolation and access policies with Sentry monitoring from separate hardware. NVIDIA says Sentry can quarantine and stop agents that try to cross their software boundaries. Scale AI is using the reference design for enterprise and government applications; no deployment outcomes or independently verified effectiveness are reported.

People

Vendors

Products

Industries

Related Articles