Nvidia says its new AI safety platform can contain rogue agents within ‘milliseconds’

Nvidia has launched its Open Agent Safety Platform to contain and monitor AI agents, responding to recent rogue hacking incidents. The platform, utilizing Nvidia’s OpenShell open-source software and Vera AI CPU, enforces boundaries by checking access restrictions before and during tasks. It also incorporates Nvidia’s Sentry technology on a separate chip for continuous monitoring. CEO Jensen Huang emphasized providing AI agents with minimal necessary rights, and major tech companies like Anthropic, Microsoft, and SpaceX are backing the platform.

Unlike previous approaches, Nvidia's platform uses a separate chip for continuous monitoring and claims to quarantine rogue agents within milliseconds.