NVIDIA Introduces Hardware Backed Agent Safety Platform

NVIDIA introduces the Open Agent Safety Platform, a reference design for monitoring and controlling AI agents, as detailed in a company blog post. The platform combines OpenShell on NVIDIA Vera CPUs with NVIDIA Sentry on BlueField-4 data processing units.

OpenShell is an open source runtime that places agents in isolated sandboxes. Operators can define which files, networks, tools, processes, and credentials an agent may access, then verify and enforce those limits while the agent runs.

NVIDIA Sentry adds an independent monitoring and enforcement layer using BlueField hardware and NVIDIA DOCA. In Vera Rubin POD systems, BlueField-4 sits on the node's path to the model, where it can record agent activity and enforce policies in real time without giving the agent access to those controls. The platform also supports other hardware systems.