NVIDIA Open Agent Safety Platform
Open, customizable tools that enforce boundaries around AI agents.
Overview
Engineering Full-Stack Safety
As agents become a digital workforce, developers need tools to verify the safety of their products. Without the right engineering solutions, AI agents can take actions with unintended consequences. NVIDIA Open Agent Safety Platform is an open reference design built with partners that continuously monitors and governs agent behavior, ensuring that AI agents follow the rules.
It features NVIDIA OpenShell™, an open source runtime that provides governance tools to enforce what an agent can see, do, and interact with. The controls remain in force when agents behave unexpectedly, limiting how far mistakes spread. For organizations that want an additional, independent layer, NVIDIA Sentry provides out-of-band, in-silicon telemetry of agent activity and security-policy enforcement, enabling it to quarantine agents in milliseconds.
All of this is optimized to run on NVIDIA Vera CPU and NVIDIA BlueField DPU-based systems, and is also compatible with other hardware systems.
Benefits
A Layered Foundation for Agent Safety
Combine runtime governance, continuous threat detection, and hardware-isolated policy enforcement to keep AI agents contained, observable, and auditable.
Technology
Explore NVIDIA Open Agent Safety Platform
The software and hardware that govern, protect, and power AI agents at enterprise scale.
Ecosystem
Leading Adopters Across All Industries
Industry leaders from across the AI ecosystem are joining NVIDIA to strengthen AI safety for every industry across the full stack of infrastructure, software, models and robotics.
Next Steps
FAQ
Each component serves a distinct role:
- NVIDIA OpenShell governs how agents execute, what they can access and change, and where inference runs.
- Sentry with BlueField-4 delivers tenant isolation outside the agent’s execution environment and provides in-silicon threat detection, identity governance, and policy enforcement using DOCA.
- NVIDIA Vera CPUs provide high-performance, power-efficient compute for orchestration, sandboxed code execution, and data processing.
Learn more about NVIDIA Cybersecurity and NVIDIA AI Security Research.
Prompts, model safeguards, and agent frameworks influence what an agent attempts to do. Runtime controls enforce what it is allowed to do. OpenShell applies policy outside the agent process, while NVIDIA Sentry with NVIDIA BlueField-4 adds an independent security layer outside agent and host software.
Yes. OpenShell supports agents such as Claude Code, Codex, OpenCode, GitHub Copilot CLI, and OpenClaw. Teams can also bring custom agents and sandbox images. OpenShell supports open and closed models and provides a common runtime policy layer across agent workflows.
No. OpenShell can run on supported local, on-premises, cloud, and Kubernetes infrastructure without BlueField-4. On systems with BlueField-4, Sentry adds hardware-isolated monitoring and enforcement that remain operational if the host or workload is compromised.
OpenShell provides an audit trail of allow and deny decisions and supports centralized collection of sandbox logs. Agents can request policy changes, with optional automatic approvals constrained by approved policy limits and enterprise governance. Operators retain control over the boundaries agents must follow.