Securing the Next Generation of AI Infrastructure at Runtime

Artificial intelligence is rapidly transforming the enterprise landscape, powering everything from autonomous agents to large-scale LLM applications. However, as organizations adopt AI infrastructure at scale, they face an urgent challenge: ensuring the integrity, safety, and trustworthiness of their AI operations in the face of increasingly sophisticated cyber threats. Moreover, a new set of threats comes to mind as new attack surfaces are created.

As AI becomes the operational backbone of modern enterprises, CISOs and security engineering leaders are facing a dual challenge:

1. AI workloads demand exceptional performance and trust.
2. They also introduce entirely new attack surfaces, failure modes, and dependency chains.

For this reason, we’re excited to announce upwind is leveraging NVIDIA technologies that directly address both problems, bringing runtime-first security to the heart of accelerated AI computing.

Why This Matters for Security Leaders

AI infrastructure is an ecosystem of GPU accelerators, inference services, orchestration layers, and model-specific pipelines, each introducing unique risks.

Upwind is unifying performance, visibility, and protection across AI environments. Using nvidia technologies , we provide runtime-first security engineered directly into the systems that enterprises rely on to run AI at scale.

Centering on two capabilities from NVIDIA:

• NVIDIA NIMs (NVIDIA Inference Microservices): powering Upwind’s internal AI-driven security operations, from runtime analytics to large-scale vulnerability correlation and threat modeling.
• NVIDIA Garak: integrated into Upwind’s LLM security validation layer for adversarial testing, jailbreak simulation, and data exfiltration detection.

Together, these capabilities enable continuous validation of AI applications, backed by real runtime context, workload behavior, and API observability from the Upwind platform. In addition to leveraging NVIDIA technology for AI-driven security operations, Upwind also now provides robust security for NVIDIA AI workloads.

“As AI takes on a central role in business and infrastructure, developers must design systems that are secure from the start. By incorporating NVIDIA’s accelerated computing, advanced AI frameworks, and security-ready infrastructure, Upwind is changing how organizations understand and defend modern cloud environments.”

-Ariel Levanon, Vice President of Cybersecurity, NVIDIA

Securing GPU-Powered AI Workloads

Security leaders are already preparing for a world where AI systems are targeted with model poisoning, inference manipulation, supply chain compromise, or runtime exploitation. This collaboration directly addresses these high-impact scenarios.

Upwind now provides dedicated protection for AI workloads running on NVIDIA GPU-based systems, including the NVIDIA DGX™ platforms and Blackwell™ architecture. That includes:

• Continuous runtime visibility across GPU-powered environments
• Real-time risk prioritization tailored to AI infrastructure
• Zero performance impact on inference or training pipelines

This is powered by an upwind framework tailored for the Nvidia AI suite and delivers five core advantages:

1. Enhanced performance through GPU acceleration
2. Deployment flexibility across sovereign and private clouds
3. Cost-efficient scalability for inference and analytics
4. Strict data privacy and locality enforcement
5. Tailored engineering aligned to customer-specific AI architectures

As a result, security teams gain a real-time understanding of how AI workloads behave, not just the configuration state. And when an abnormal API call, inference pattern, or process execution emerges, Upwind can surface it instantly with the full runtime context needed to take action.

AI-Driven Security Powered by NVIDIA NIM microservices

NVIDIA NIM microservices are now integrated directly into both Upwind’s platform architecture and our internal engineering workflows, serving as the foundation for deeper runtime context and AI operations.

NIM microservices provide a standardized, containerized framework for deploying high-performance NVIDIA models across any environment: public cloud, private cloud, on-prem GPU clusters, or sovereign regions. By adopting NIM microservices, Upwind can deliver advanced AI-driven security capabilities while remaining cloud-agnostic, operationally flexible, and aligned with the strict compliance requirements of global enterprises.

Within Upwind’s runtime security engine, NIM microservices accelerate the AI components that matter most for real-time protection, such as contextual inference, anomaly analysis, automated reasoning, and correlation across cloud assets, data flows, workloads, and GPU-powered AI systems. This gives security teams the ability to cut noise, detect meaningful signals, and surface advanced threats.

NIM microservices also power Upwind’s internal AI operations ranging from product research and data processing pipelines to secure model evaluation. By building on standardized NVIDIA components, Upwind can iterate faster, maintain consistent performance across environments, and support stronger data sovereignty controls for customers who require local model execution.

The result is a unified, AI-driven security layer that brings NVIDIA-grade inference and intelligence directly into cloud and AI infrastructure. With NIM microservices, customers gain:

• High-performance AI inference at scale for cloud and GPU environments
• Consistent, portable model deployment across sovereign and regulated workloads
• Enhanced runtime context and automated reasoning for precise risk reduction
• Cloud-agnostic architecture with no vendor lock-in
• Stronger data protection and governance through local model execution

By embedding NVIDIA NIM microservices into the Upwind platform, we’re not just accelerating AI, we’re redefining how enterprises secure cloud and AI systems in real time.

LLM Safety and Security Validated in Real Time

The integration of NVIDIA Garak strengthens one of the most important gaps emerging in enterprise AI programs: validation of LLM robustness against adversarial behavior.

With Garak, Upwind continuously exercises models against attacks such as:

• Prompt injection
• Jailbreak attempts
• Data extraction or exfiltration
• Manipulation of model behavior

Combined with Upwind’s runtime telemetry and API observability, security teams gain a feedback loop for model integrity and “under stress” activities that reflect real-world behavior.

“NVIDIA is setting the foundation for enterprise AI, and we’re proud to both leverage it and secure it. By combining Upwind’s runtime visibility and protection with NVIDIA’s accelerated AI infrastructure, we’re helping organizations deploy AI at scale—safely, efficiently, and with full confidence. Leveraging NVIDIA core technologies such as NVIDIA NIM microservices and NVIDIA Garak, Upwind uses AI to deliver exceptional security outcomes for our customers.”

-Dan Yahav, SVP Platforms, Upwind

Raising the Standard for Trusted AI

This is part of Upwind’s broader AI security strategy, which includes:

• AI workload runtime protection
• AI vulnerability management
• LLM-aware API security
• AI-SPM

Upwind , using NVIDIA, is moving security closer to where AI actually run on real workloads and GPU systems, in real time. To learn more about how Upwind both leverages AI within its platform and also secures AI infrastructure, schedule a demo today.