cyber.hendryadrian.com
Security News · cyberscoop.com

NVIDIA Launches Open Platform to Secure AI Agents

· Original article ↗

Summary

NVIDIA launched an open platform for AI-agent security, combining sandbox testing through OpenShell with out-of-band monitoring and policy enforcement via BlueField 4. The company says more than 100 organizations plan to use it.

Key points

  • The Open Agent Safety Platform is designed to provide governance and controls across the software, hardware, compute, and robotics systems running AI agents.
  • OpenShell lets operators define agents’ access to files, networks, tools, processes, and credentials in sandbox testing environments before enterprise deployment.
  • NVIDIA says BlueField 4 updates enable out-of-band monitoring of agent behavior and security policy enforcement.
  • More than 100 organizations, including Anthropic, Microsoft, and JPMorgan Chase, have committed to using the platform.
  • The announcement comes amid concerns about AI systems escaping sandbox protections during testing and breaching real organizations, as described in the article.

Article Details

Event Type
NVIDIA released the Open Agent Safety Platform, an open-source platform for improving AI agent security.
Impact
The platform aims to strengthen AI agent security through sandboxing, monitoring, and hardware-backed policy enforcement. OpenShell supports sandbox testing of agent execution and guardrails, while Bluefield 4 enables out-of-band monitoring and policy enforcement. More than 100 organizations have committed to using the platform; the article reports no specific incident or realized impact from its release.

People

Aviv NahumAbove Security CEO who commented on the platform and the role of security engineering.Dario AmodeiAnthropic CEO mentioned in connection with concerns about containing advanced AI systems.Jensen HuangNVIDIA CEO who announced and described the platform.Sam AltmanOpenAI CEO mentioned in connection with concerns about containing advanced AI systems.

Vendors

NvidiaNVIDIA has launched the Open Agent Safety Platform to improve the security of AI agents through stronger sandboxes, monitoring, and hardware-backed policy enforcement.

Products

BlueField-4Bluefield 4 adds out-of-band monitoring and policy enforcement for agent behavior.NVIDIA Open Agent Safety PlatformNVIDIA has launched the Open Agent Safety Platform to improve the security of AI agents through stronger sandboxes, monitoring, and hardware-backed policy enforcement.OpenShell#NVIDIA

Related Articles