日本語 
Akamai is partnering with NVIDIA to bring a sophisticated security architecture to AI factories, which will form the foundation for the agentic era as AI agents become more widespread.
Akamai and NVIDIA have announced an expansion of their security collaboration, integrating Akamai Guardicore Segmentation into the NVIDIA Vera BlueField-4 STX storage architecture , which is based on the NVIDIA DOCA software platform . This collaboration aims to protect data, context memory, and autonomous agents by embedding a zero-trust architectural layer into the AI factory itself.
This groundbreaking security integration allows AI factory operators to apply workload-aware segmentation, monitor agent behavior, and contain threats at the infrastructure layer, all while maintaining the high speed of accelerated computing without burdening the processing cycles of the GPUs, CPUs, or storage on which AI workloads depend.
Also Read: Exabeam Launches AI Agent Security Verification Tool
Security achieved through high-speed processing
The AI factory is being built at such a rapid pace that security measures can’t keep up. Previously, there was a trade-off between AI speed and AI security. However, such compromises are no longer acceptable.
The expanded integration between Akamai and NVIDIA is based on the architectural agreement announced by both companies in February of this year and is designed to resolve the trade-offs involved.
Akamai Guardicore Segmentation protects the world’s largest and most sensitive organizations, providing an intelligence layer that continuously maps how workloads, applications, and data interact across hybrid environments, including data centers, cloud infrastructure, Kubernetes clusters, and edge systems. Policies are defined based on workload identity, application context, and runtime behavior, rather than static network addresses. By providing visibility into the entire lifecycle of AI workloads, it uncovers anomalous patterns and unauthorized access to sensitive information.
The NVIDIA Vera BlueField-4 STX, programmable via NVIDIA DOCA, builds threat detection and enforcement layers on the silicon. Security policies are applied at line speed to data paths within the infrastructure fabric, rather than to hosts, ensuring policy enforcement occurs closer to the workload itself. This does not hinder the processing of the GPUs, CPUs, and storage processors on which the AI factory depends. The two layers work together to establish identity-based zero trust as an inherent characteristic of the infrastructure itself, rather than an add-on product.
SOURCE: PRTimes
English