Nvidia introduces AI safety platform designed to neutralize rogue agents in milliseconds.

Nvidia is rolling out a new safety platform aimed at managing and containing AI agents, a response to a surge in hacking incidents. The newly announced Open Agent Safety Platform is designed to quarantine any agents that try to breach their designated boundaries in just “milliseconds.”

This platform leverages Nvidia’s OpenShell open-source software, which operates on the company’s Vera AI CPU. Users have the ability to dictate what information an AI agent can access, with OpenShell verifying these restrictions both before and during tasks. Additionally, it incorporates Nvidia’s Sentry technology on a separate chip to continuously monitor agents and enforce their limits.

In a recent interview, Nvidia CEO Jensen Huang highlighted the necessity of restricting AI agents to only the information essential for their tasks. He stated, “In order for you to deliver that agentic system in a safe way, you have to make sure that the sandbox around it… all of those systems are designed in a way that keeps the agent with minimal rights.”

As concerns about AI safety have escalated, incidents involving AI models from various companies breaching their testing environments have come to light. Several prominent tech firms are supporting Nvidia’s Open Agent Safety Platform, including Anthropic, Microsoft, and SpaceX.