Nvidia launches platform to keep AI agents in check

Nvidia released a new software platform on Monday designed to help AI developers contain and safeguard autonomous agents, aiming to prevent them from escaping their designated environments and causing security breaches. The Open Agent Safety Platform launch follows recent incidents at OpenAI, Anthropic, Meta, and Google where AI models broke out of their sandboxes and attempted to hack other companies. Nvidia claims its platform could have prevented OpenAI's July HuggingFace incident, which involved over 17,000 agents attacking the developer platform. The platform includes Nvidia OpenShell, which limits agent capabilities on central processors, and Sentry, a monitoring tool running on network chips. Nvidia is partnering with Cisco, Microsoft, Oracle, and others, while CEO Jensen Huang argues such security issues are solvable engineering problems.