Supermicro Supports NVIDIA Open Agent Safety Platform: From Policy to Hardware-Backed Enforcement

Extending familiar enterprise security principles across the agent runtime and AI infrastructure — bringing established practices to a new execution model.

Agentic AI introduces a new class of workload, but the security principles are familiar. NVIDIA Open Agent Safety Platform is an open software platform and reference system design to secure agents from testing through deployment. NVIDIA OpenShell brings least-privilege, policy-based controls to the agent at runtime, while NVIDIA Sentry on NVIDIA BlueField^®-4 uses NVIDIA DOCA for out-of-band monitoring and policy enforcement in a hardware-isolated infrastructure domain Supermicro helps customers bring these layers together on proven AI infrastructure.

Extending enterprise-grade security to agentic AI

For more than three decades, Supermicro has helped enterprises deploy data center infrastructure where security, reliability, and operational control are fundamental requirements. That experience includes platform capabilities such as hardware root of trust, secure boot, confidential computing, memory protection, and DMA protection. As autonomous agents gain access to enterprise data, tools, APIs, and long-running workflows, those same trusted infrastructure principles need to extend to a new execution model.

Recent incidents involving frontier agents reinforce that AI safety is a legitimate concern and that applying established enterprise security principles to this new execution model is a shared responsibility. Model developers, infrastructure providers, deployers, and users must each address the risks within their control. Together, these layers form a familiar security model that enables enterprises to adopt increasingly autonomous agents with greater confidence.

NVIDIA OpenShell provides the secure runtime boundary around the agent, while NVIDIA Sentry, running on NVIDIA BlueField-4 DPU, verifies agent identity, enforces zero-trust access to data, tools, APIs, and services, and provides attested telemetry to inspect agent behavior and detect deviations. If an agent attempts to move outside its permitted boundaries, Sentry can quarantine it in milliseconds, helping contain unauthorized activity. NVIDIA Vera CPUs provide high-performance compute for CPU-intensive agent workloads such as orchestration, sandboxed code execution, tool use, and data processing.

Supermicro and NVIDIA: aligned on high-performance, secure AI

Aligned with the NVIDIA Open Agent Safety Platform, Supermicro has the shared goal of delivering high-performance and secure AI products and solutions. Supermicro tested early versions of OpenShell across Supermicro hardware platforms, helping prepare for broader enterprise adoption of autonomous AI agents.

Supermicro also brings a significant installed base of NVIDIA-based AI systems deployed across neocloud providers and enterprise customers worldwide. OpenShell runs across cloud, hybrid, on-premises, Kubernetes, and other supported runtime environments. This gives customers a practical way to introduce agent-level safeguards on supported infrastructure rather than treating agent safety as an immediate rip-and-replace transition.

NVIDIA BlueField and DOCA extend enforcement beyond the agent

NVIDIA BlueField-4 provides a hardware-isolated, host-independent trust domain for security services, separate from the agent’s execution environment. Its dedicated Grace CPU cores run security software, while hardware accelerators offload supported networking and security operations. Within this domain, Sentry uses DOCA programmable security capabilities to provide out-of-band monitoring, policy enforcement, and attested telemetry. DOCA enables runtime inspection of host memory to detect unauthorized execution and changes in workload integrity, alongside inspection of agent traffic and enforcement of network and application-level access policies. These capabilities help identify suspicious behavior and restrict unauthorized connections without depending on the agent to report its own activity. These controls remain operational even if the host or workload is compromised, giving operators visibility and control without relying on agent-reported logs or host-based security software. BlueField-4’s dedicated processing resources also offload these security services from host CPUs, preserving host compute for agent workloads.

Supermicro has extensive experience integrating NVIDIA BlueField technology across AI, networking, and storage platforms.

OpenShell adds a secure runtime boundary

OpenShell governs how an agent executes, what it can see and do, and where inference goes. It applies least-privilege policy around files, processes, network access, credentials, and model endpoints while working with open or closed models and a range of agent harnesses. For enterprise customers, the value is straightforward: extend familiar zero-trust and policy-based controls to autonomous agents without treating every deployment as a greenfield infrastructure project.

Protecting existing AI investments while building for the future

Supermicro's broad NVIDIA portfolio gives customers flexibility in how they deploy agent security. Organizations can begin with NVIDIA OpenShell on supported existing Supermicro NVIDIA environments, and add NVIDIA Sentry on NVIDIA BlueField-4 where platform support and validation allow. At the same time, Supermicro can deliver next-generation NVIDIA Vera Rubin NVL72, NVIDIA HGX Rubin NVL8, storage, high-speed networking, liquid cooling, and Data Center Building Block Solutions® (DCBBS) for customers building complete AI factories. Supermicro combines engineering excellence with speed to market, helping customers rapidly configure, integrate, and validate NVIDIA-based systems for secure agent workloads.

Agent safety does not have to begin with a complete infrastructure replacement: OpenShell creates a path to extend safeguards across supported existing environments while customers evolve toward deeper hardware-isolated enforcement.

A practical path to customer adoption

With NVIDIA OpenShell available for enterprise production and adoption, Supermicro will work with NVIDIA to support deployment on validated customer environments. Customers can start with agent-level runtime safeguards on appropriate current infrastructure, introduce additional NVIDIA Sentry on BlueField-4enforcement where supported, and move toward Vera CPU and BlueField-4 platforms when they need NVIDIA's newest full-stack architecture.

For customers, this approach provides a clearer path from agent experimentation to govern production: preserve existing infrastructure investments where practical, extend consistent security principles across multiple deployment environments, and add stronger controls as autonomous agents take on more business-critical work.

Building safe and responsible AI together

Supermicro will continue to support evolving requirements for AI security, safety, governance, and compliance, working across the broader technology ecosystem to enable safe and responsible AI deployment. Together with NVIDIA, Supermicro is helping customers extend proven enterprise security practices into agentic AI.