Efficient Micro-Segmentation Generation for Wireless Network Zero-Trust Security: A Graph Diffusion-Based Approach

Zero-trust security has emerged as a critical form for addressing the unique vulnerabilities of wireless networks, but its implementation faces significant practical challenges. First, realizing zero-trust requires partitioning the network into multiple isolated, application-specific micro-segmentations with customized security policies. Second, to accommodate increasingly complex service demands, service provisioning within micro-segmentations should utilize Service Function Chains (SFCs) that distribute services across heterogeneous devices. Therefore, this paper proposes an efficient micro-segmentation generation approach called LLM-enhanced Graph Diffusion (LGD). Specifically, we model zero-trust wireless networks as a hierarchical graph structure that captures both physical characteristics and trustworthiness relationships, and formulate the micro-segmentation generation problem as a controllable generation problem. Additionally, we present LGD based on graph diffusion models that optimize micro-segmentation through a progressive denoising process. Furthermore, LGD leverages the cognitive capabilities of Large Language Models (LLMs) to reduce action space dimensions through intelligent filtering and heuristic guidance, thereby accelerating convergence and improving solution quality. Extensive experiments demonstrate that LGD-generated micro-segmentations achieve 40% higher service provisioning efficiency compared to existing baseline approaches. Moreover, comprehensive evaluations across various network scenarios showcase the scalability and robustness of the proposed approach.