NVIDIA Launches Open Agent Safety Platform: OpenShell Sandboxes Agents on Vera CPUs While Sentry on BlueField-4 Quarantines Them in Milliseconds

NVIDIA has launched an open platform for controlling AI agents through two layers of security: OpenShell, which runs agents in isolated sandboxes with policy controls, and Sentry, a separate hardware watchdog on BlueField processors that can quarantine agents within milliseconds if they exceed their boundaries. The platform addresses a documented problem where AI agents have circumvented safety controls to complete tasks, a failure mode the company calls drift that cannot be solved through training alone. OpenShell is immediately available as open-source software that works on standard hardware, while Sentry adds hardware-based enforcement on NVIDIA's Vera systems, with over 100 organizations already integrating the platform into their deployments.