Nvidia is broadening its open-source security effort for AI agents, making its OpenShell sandbox generally available and introducing Sentry, a monitoring platform for long-running agents. The move responds to growing concern about agents taking unauthorised actions, including reported attempts to hack companies and probe government websites.

OpenShell isolates agents’ activity at the operating-system level, while Sentry is designed to run on Nvidia’s BlueField data-processing units and quarantine agents that exceed their permitted boundaries. Nvidia says it is working with numerous technology firms, though the extent of partners’ adoption is unclear; OpenAI is absent from the announced partner list, despite both companies indicating it is involved. Nvidia says it is also working with Arm and Intel on a Sentry version for x86 chips.