Nvidia launches a security platform for artificial intelligence agents — here’s how it works

Nvidia is launching a security platform for artificial intelligence (AI) agents that more than 100 organizations are collaborating on, including Anthropic, SpaceXAI and JPMorganChase.

Called Nvidia’s Open AI Agent Security Platform, it is designed to place limits on what agents can access and to put them in “quarantine” within milliseconds if they exceed those limits.

The platform can be used across the software and devices, computer and robotic systems that run agents.

“The extraordinary potential of artificial intelligence for society will only be realized if we solve the problem of artificial intelligence security,” said Jensen Huang, Nvidia’s CEO and founder.

“As we continue to discover the frontiers of AI capabilities, we must accelerate discovery at the frontiers of AI security… Together, we can raise the bar for global AI security.”

Nvidia’s Open Agent Security Platform consists of two tools, OpenShell and Sentry.

OpenShell runs on central processing units (CPUs), the chips that handle much of the work needed to run AI agents. Nvidia says it runs efficiently on Vera, its latest CPU designed for AI agents, and can also be adapted for chips made by other companies.

Sentry is an additional “overseer” system that will run on a separate Nvidia chip, watching agents that pass software checks.

Organizations working with Nvidia are using or developing the technology in different ways, Nvidia says.

Anthropic has partnered with Nvidia to add security controls to its Claude service for businesses to use artificial intelligence agents, allowing companies to limit these agents’ access.

SpaceX’s AI division, SpaceXAI, says it is using the platform for Cursor coding agents and Grok models, while Salesforce has integrated OpenShell with Slack so teams can see what agents are doing and approve or reject requests for additional permissions.

In the finance sector, JPMorganChase and Citi are collaborating with Nvidia on shared open-source agent security technology.