NVIDIA Launches Open Agent Safety Platform to Enforce AI-Agent Controls Across Software and Hardware
Summary
NVIDIA launches the Open Agent Safety Platform, pairing open-source OpenShell sandboxes with NVIDIA Sentry and BlueField-4 DPUs to enforce verifiable AI-agent rules on files, networks, tools, processes and credentials before and during execution.
Key Points
- NVIDIA introduces the Open Agent Safety Platform, combining open-source OpenShell with NVIDIA Sentry to monitor and enforce AI-agent safety across software and hardware.
- OpenShell runs each autonomous agent in a kernel-isolated sandbox and converts operator rules on files, networks, tools, processes and credentials into verifiable policies checked before and during execution.
- In Vera Rubin PODs, BlueField-4 DPUs sit on each node’s only path to the model, enabling out-of-band policy enforcement and observability at line speed; existing Vera and BlueField-4 users can enable the protections through a software update.