Integration with NVIDIA DOCA security and NVIDIA Vera BlueField-4 STX will enable Xage to deliver line-speed policy enforcement and visibility at AI factory scale for models, data, agents, and context memory.
AI agents are increasingly the ones making decisions, not just assisting with them. They query databases, trigger workflows, call APIs, and interact with infrastructure that was never designed with autonomous software actors in mind. The security industry has spent years building perimeter defenses, endpoint controls, and access management tools around human users. The question those tools were never designed to answer is now becoming urgent: when an AI agent requests access to a sensitive system, what actually governs what it can do once it gets in?
Xage Security and NVIDIA are working toward a concrete answer. Xage has announced support for NVIDIA’s newly unveiled DOCA security capabilities and NVIDIA Vera BlueField-4 STX, integrating its Zero Trust architecture directly into the silicon layer where AI workloads execute. The combination is designed to give enterprises continuous visibility and policy enforcement across AI factory environments, without requiring changes to host software or the workloads themselves.
What NVIDIA Vera BlueField-4 STX Actually Does
The BlueField-4 STX is not a conventional security appliance inserted between systems. It is a data processing unit that embeds security functions into the infrastructure itself, running at line rate alongside AI workloads rather than inspecting them after the fact. NVIDIA‘s DOCA security framework provides the software layer that sits on top, enabling granular policy enforcement, real-time threat detection, and visibility into what AI agents are doing at the workload level.
The architecture matters because it sidesteps one of the persistent problems in securing AI factory environments: adding security controls that degrade the performance of the AI systems they are protecting. Enforcement at the silicon boundary means the governance layer runs in parallel rather than in series with the workload.
Where Xage Fits Into That Architecture
Xage brings the identity and policy intelligence that BlueField-4 STX’s enforcement infrastructure needs to make access decisions. The integration works across several layers. Xage Security Gateways run natively within the NVIDIA DOCA architecture using the Xage Resource Gateway, Agent Sentry, and Extended Protection components, covering both resources and the agents trying to reach them. The Xage Resource Gateway integrates with DOCA Vault for file access visibility and control. The Xage Fabric’s policy engine draws from DOCA data in real time to evaluate interactions as they happen rather than auditing them afterward.
For threat detection, Xage feeds identity-level intelligence into DOCA Argus and DOCA Flow, sharing events including login attempts, entitlement delegation, and credential changes, giving BlueField-4 STX’s detection capabilities a richer context to work from. Hardware-accelerated enforcement runs through DOCA OvS, where Xage dynamically configures the stack to permit authorized interactions and block anything that falls outside policy at line speed.
The result is a closed-loop model: DOCA observes runtime behavior at the infrastructure level, Xage evaluates identity, policy, resource context, and the action being requested, and enforcement controls act through DOCA-OvS before unauthorized activity creates downstream exposure.
The Problem This Is Actually Solving
The case for this architecture starts with a gap that most enterprise AI deployments are currently ignoring. Agentic AI systems move across APIs, SaaS platforms, databases, cloud services, file storage, and internal applications, often with access provisioned broadly enough to function across varied environments. The access control models most enterprises have are designed around human users with relatively predictable behavior patterns. AI agents do not follow those patterns. They can query systems at high volume, chain together actions across multiple environments in seconds, and operate continuously without the session boundaries that make human access easier to monitor.
Duncan Greatwood, CEO of Xage Security, framed the core requirement plainly: “As autonomous AI agents gain access to sensitive data, APIs, applications, and core systems, organizations need unbypassable visibility into and control over what those agents can see, do, and change.”
The word unbypassable is doing real work in that statement. Software-based security controls that run on the same host as the workloads they are protecting can, under certain conditions, be circumvented by a compromised or misbehaving agent. Enforcement baked into the silicon layer is structurally harder to route around.
Governance Across the Full Interaction Chain
Most approaches to AI agent security focus on either the prompt layer or the model output. Xage’s architecture operates across the full interaction chain, covering users, agents, models, tools, APIs, applications, infrastructure, and resources. Least-privilege access, just-in-time and just-enough policy enforcement, lateral movement blocking, and automated threat response are all part of the same governance surface rather than separate tools that need to be coordinated manually.
For enterprises, government agencies, and critical infrastructure operators that are moving agentic AI from controlled pilots into environments where the stakes of an unauthorized action extend beyond data, that breadth of coverage is what makes the difference between a security posture that holds and one that creates gaps exactly where the pressure is highest.
The AI factory is becoming foundational infrastructure. Xage and NVIDIA’s integration is a direct argument that it should be secured like it.
Research and Intelligence Sources: Xage Security, NVIDIA
To participate in our interviews, please write to our CyberTech Media Room at info@intentamplify.com
