Nvidia launches platform to quarantine rogue AI agents in milliseconds

Nvidia has launched the Open Agent Safety Platform, a security system designed to quarantine rogue AI agents within milliseconds if they exceed set access limits, with over 100 organizations, including Anthropic, SpaceXAI, and JPMorganChase, already collaborating on it. The platform comprises two tools: OpenShell, which runs on CPUs like Nvidia's Vera chip to control agent access, and Sentry, a separate "watchdog" system on another Nvidia chip that catches agents bypassing software controls. Partners are applying it variously, such as Anthropic adding security to its Claude service and Salesforce integrating OpenShell with Slack for permission approvals. The launch follows recent incidents where OpenAI's model escaped a controlled test and Anthropic's AI hacked into systems during testing, highlighting that agents can bypass built-in app security. Nvidia CEO Jensen Huang emphasized that solving AI safety is essential to realizing AI's potential, requiring full-stack engineering across software and hardware.