NVIDIA Launches Open Agent Safety Platform to Secure Agents From Testing to Deployment
Short summary
NVIDIA released an open platform combining software (OpenShell) and hardware (Sentry on BlueField DPUs) to monitor and control AI agents, preventing them from breaking out of assigned boundaries. Over 100 companies including Anthropic, Microsoft, and SpaceX are adopting it to enforce security policies across the full stack.
- OpenShell provides a secure runtime boundary on NVIDIA Vera CPUs that traces agent actions and enforces policies, with the ability to run on third-party chips from Arm and Intel as open source software.
- Sentry runs as an independent watchdog on BlueField-4 DPUs and can quarantine agents in milliseconds if they attempt to exceed their boundaries, operating outside the agent's reach.
- Recent agent security incidents showed attackers circumventing application-layer controls, so this platform moves enforcement to hardware and infrastructure layers where agents can't bypass it.
Full summary
NVIDIA announced an open-source safety platform for AI agents that operates across three layers: software, hardware, and robotics systems. The core components are OpenShell, a secure runtime that sets boundaries for agents running on NVIDIA Vera CPUs, and Sentry, a watchdog system on BlueField-4 data processing units that monitors agent behavior independently in silicon. The problem they're solving is concrete — recent security incidents show agents circumventing application-layer controls to complete tasks, so these tools enforce restrictions at lower levels where agents can't bypass them. OpenShell works with third-party chips from Arm and Intel too, making it not vendor-locked.
Sentry operates in real time, quarantining rogue agents within milliseconds by monitoring requests and responses, verifying agent identity, and enforcing zero-trust access policies. It runs isolated from the agents themselves, meaning attackers can't disable it from within the system. The platform includes integration points — Anthropic's Claude Managed Agents separates the agent loop from execution sandboxes, Salesforce embedded controls into Slack for team oversight, and SAP is pairing it with their business AI platform. SpaceX is using it for Cursor coding agents and Grok models, while robotics companies like Figure and Skild AI are embedding safety controls into physical systems.
The adoption list reads like a tech industry roster: over 100 organizations are working with the platform, including Microsoft, Palantir, CrowdStrike, JPMorgan Chase, and major infrastructure providers like Hitachi Energy and Schneider Electric. Red Hat, Canonical, and SUSE are integrating it into operating systems. Scale AI is building it into their agentic infrastructure for enterprise and government customers. The approach reflects Jensen Huang's framing that "safety requires full-stack engineering" — you can't just trust the model layer, you need controls baked into hardware and infrastructure too.