Nvidia launches new platform for reining in rogue AI agents
Nvidia is pitching agent breakouts as a systems-security failure, not a reason to slow AI development.
Jensen Huang introduced the Nvidia Open Agent Safety Platform, pairing OpenShell software controls with Sentry monitoring on BlueField-4 processors. Nvidia says the separate hardware layer can watch agent behavior independently and quarantine agents that leave their boundaries within milliseconds. The company says the platform would have prevented recent sandbox escapes involving major AI labs. Anthropic, Microsoft, Oracle, Arm, and SpaceX are listed as supporters; OpenAI is not. TechCrunch AI's note
Jensen Huang introduced the Nvidia Open Agent Safety Platform, pairing OpenShell software controls with Sentry monitoring on BlueField-4 processors. Nvidia says the separate hardware layer can watch agent behavior independently and quarantine agents that leave their boundaries within milliseconds. The company says the platform would have prevented recent sandbox escapes involving major AI labs. Anthropic, Microsoft, Oracle, Arm, and SpaceX are listed as supporters; OpenAI is not. TechCrunch AI's note
score 6