Nvidia Launches Open Agent Safety Platform to Contain Rogue AI Agents

Nvidia has introduced a new platform designed to keep AI agents confined to approved environments using independent software and hardware security controls.

CEO Jensen Huang announced the Nvidia Open Agent Safety Platform on Monday following several incidents in which AI models reportedly bypassed safeguards and accessed systems outside their intended testing environments.

“AI’s extraordinary potential for society will only be realised if we solve AI safety,” Huang said, describing safety and security as a full-stack engineering challenge.

OpenShell and Sentry Add Separate Lines of Defense

The platform combines OpenShell, Nvidia’s open-source software for controlling what AI agents can access, with an independent monitoring system called Sentry.

OpenShell establishes software boundaries around an agent, while Sentry runs separately on Nvidia’s BlueField-4 data processing units rather than the CPU or GPU operating the agent. Nvidia says this gives Sentry an isolated view of activity and allows it to quarantine agents within milliseconds if they move outside permitted boundaries.

OpenShell was originally announced in March. Nvidia is now combining it with hardware-level monitoring to strengthen enforcement around agent deployments.

Nvidia Treats Agent Safety as an Engineering Problem

Nvidia argues that recent AI agent breakouts can be addressed through stronger runtime environments and tighter access controls rather than slowing AI development.

Huang compared the approach to corporate security policies, where employees and executives receive only the permissions needed for their work. He said AI agents should similarly begin with limited rights and gain access only when necessary.

Nvidia said dozens of technology companies support or use the open-source effort, including Anthropic, Arm, Microsoft, Oracle and SpaceX.

The company said the broader effort began about a year ago following the introduction of OpenClaw, an operating system for AI agents created by Peter Steinberger. Nvidia later launched NemoClaw, its enterprise-focused agent platform with built-in security controls.

AI Agent Security Debate Continues

David Sacks supported the engineering-focused approach in a post on X, arguing that recent incidents reflected weaknesses in the systems surrounding AI agents.

“Recent breakouts weren’t proof that development must stop,” Sacks wrote. “They were proof that the sandbox was too weak. The runtime environment was poorly designed and misconfigured.”

Nvidia is betting that combining software restrictions with independent hardware monitoring can provide stronger containment as AI agents gain access to more powerful tools and real-world systems.