Nvidia launches Open Agent Safety Platform with hardware watchdog

Nvidia has announced the Open Agent Safety Platform, combining open source software with a reference system design to keep AI agents inside set boundaries from testing through deployment. The platform has two parts: OpenShell, an open source runtime that sandboxes agents and enforces policy on their actions, and Sentry, a watchdog that runs separately on Nvidia’s BlueField-4 data processing units.

OpenShell, introduced in March and now at version 0.1.0, supports agents including Codex, Claude Code, Pi and Hermes. It uses a gateway to manage sandbox lifecycles and policies, kernel-level controls on filesystem and process activity, and a supervisor that checks outbound requests against policy. API keys are substituted outside the agent workload, and agents can propose policy changes but cannot approve their own requests.

Sentry provides out-of-band enforcement on the DPU, which operates separately from the agent’s host, so Nvidia says it can quarantine an agent that moves outside its software boundary in milliseconds even if the host is compromised. It is built on Nvidia’s DOCA software and enforces zero-trust access policies for data, tools, APIs and services.

Nvidia said more than 100 organisations are working with the platform’s technologies. Anthropic has integrated its Claude Managed Agents with OpenShell and BlueField, SpaceXAI is using it for Cursor coding agents and Grok models, Salesforce has integrated OpenShell with Slack, and SAP is embedding it in its Joule Studio runtime. CrowdStrike, Palo Alto Networks and Cisco are also among the partners.