%20(1).png)
Nscale backs security for agentic AI, with NVIDIA OpenShell and Sentry on BlueField-4 providing continuous, in-silicon protection across the Vera Rubin platform.
As AI agents move into production, their access to sensitive data and critical systems raises the stakes for security and compliance. According to Nscale’s State of AI Natives 2026 research, more than 3 in 4 AI Natives report high levels of concern about regulatory compliance. To help address these challenges, Nscale is proud to support the NVIDIA Open Agent Safety Platform.
The platform combines open software and a reference system design to secure agentic AI from testing through production on NVIDIA Vera systems.
As enterprises move AI agents into production, those agents are gaining autonomy to access data, call tools, and take action across systems. Securing them requires protection that extends beyond the models and software running them.
Built on the NVIDIA Vera Rubin platform, the Open Agent Safety Platform enforces security at the infrastructure level, delivering continuous, hardware-enforced protection that agents cannot override.
What the platform includes
- NVIDIA OpenShell is an open-source runtime that sets a secure boundary for agents, traces their actions, and enforces policy. It runs on NVIDIA Vera CPUs with minimal overhead, works across open and closed models, and is available now on GitHub.
- NVIDIA Sentry on BlueField-4 is a reference system design with an out-of-band watchdog that runs on NVIDIA BlueField-4 DPUs. Sentry monitors agent behavior, verifies agent identity, and enforces zero-trust access to data, tools, APIs, and services. It can quarantine an agent that tries to leave its boundary within milliseconds.
- NVIDIA DOCA is the software foundation for Sentry. It provides programmable inspection of agent requests and responses, plus attested telemetry.
The Open Secure AI Alliance supports the platform. NVIDIA started the alliance, which has more than 120 member organizations and is governed by the Linux Foundation.
Why it matters for Nscale customers
Nscale builds and operates AI data centers, from power and data centers to GPU compute to the software that orchestrates it. That makes infrastructure-level security a critical part of our full stack. Customers running agentic workloads on Nscale need to know their policies hold. They also need attested telemetry that proves it to regulators, customers, and their own boards.
Security delivered in silicon and out of band also means customers don’t have to choose between protection and performance. Because Sentry runs on the DPU, independently of the agent’s workload, enforcement continues without competing for the compute the agent needs.
Learn more about the NVIDIA Open Agent Safety Platform at nvidianews.nvidia.com.
.avif){kind=icon}
.png)
