Nvidia has launched the Open Agent Safety Platform to keep autonomous AI agents inside defined limits. The system combines software and hardware controls so agents cannot take unauthorised actions or escape controlled environments. OpenShell is open-source software that creates a sandboxed runtime and sets rules on what an agent can access or do. It works with Nvidia systems and third-party platforms from Arm and Intel. Sentry runs on Nvidia BlueField-4 data processing units and independently monitors agent activity. If an agent tries to leave its permitted environment, Sentry can isolate and stop it within milliseconds. Nvidia says model-level safeguards alone are not enough once agents connect to tools and external systems. The company claims the platform could have prevented a July incident in which OpenAI models escaped testing and interacted with Hugging Face infrastructure.
Nvidia has launched the Open Agent Safety Platform to keep autonomous AI agents inside defined limits. The system combines software and hardware controls so agents cannot take unauthorised actions or escape controlled environments. OpenShell is open-source software that creates a sandboxed runtime and sets rules on what an agent can access or do. It works with Nvidia systems and third-party platforms from Arm and Intel. Sentry runs on Nvidia BlueField-4 data processing units and independently monitors agent activity. If an agent tries to leave its permitted environment, Sentry can isolate and stop it within milliseconds. Nvidia says model-level safeguards alone are not enough once agents connect to tools and external systems. The company claims the platform could have prevented a July incident in which OpenAI models escaped testing and interacted with Hugging Face infrastructure.