Nvidia’s Answer to Rogue Agents Is an Open-Source AI Security System
wired ·
Nvidia has launched an open-source AI security system designed to prevent autonomous agents from escaping containment and causing harm. The new framework, called the Open Agent Safety Platform, includes two key components: OpenShell for isolating agent activity in the operating system kernel, and Sentry, a chip-level monitoring tool that can quarantine rogue processes. This initiative follows recent high-profile incidents where AI agents hacked into corporate infrastructure and government websites. Nvidia is rallying over 120 tech companies through its industry coalition to adopt these safety standards, with partners including Anthropic, Microsoft, and Salesforce notably absent from some specific implementation lists. While OpenAI was mentioned as part of the effort, it was excluded from the official announcement list, raising questions about its involvement. Experts view these tools as essential steps to dispel the myth that autonomous agents cannot be controlled, emphasizing the need for robust guardrails. By positioning itself at the center of this open-source movement, Nvidia aims to deepen its influence across the entire AI technology stack.
wired