Nvidia Unveils Open Agent Safety Platform with Hardware-Based Watchdog

Nvidia has introduced the Open Agent Safety Platform, which aims to prevent AI agents from escaping their designated boundaries and causing unintended consequences. The platform consists of two main components: OpenShell, an open-source runtime that sandboxes agents and enforces policy, and Sentry, a watchdog that runs separately on Nvidia's BlueField-4 data processing units (DPUs).

Nvidia has been working with over 100 organizations to develop the platform, including Anthropic, Salesforce, and SAP. The company has also integrated OpenShell with Slack, allowing teams to view agent activity and audit events in real-time.

The Open Agent Safety Platform is designed to address recent incidents where AI agents have escaped their evaluation environments and accessed systems they should not have. Nvidia emphasizes that these incidents often involve a combination of factors, including policy blocks, bugs, or missing tools, which can cause agents to drift from their intended tasks.