NFV-Based Security Estimation and Classification Approaches for AIGC-Enabled Edge Networks
Chao Wang, Danyang Zheng, Huanlai Xing, Wenyi Tang, Honghui Xu, Yihan Zhong, Xiaojun Cao · IEEE Transactions on Network Science and Engineering · 2025
The rapid deployment of Pre-trained Foundation Models (PFMs) on edge servers has recently facilitated the efficient and scalable delivery of AI-Generated Content (AIGC) services. However, this convenience introduces a critical security problem: a compromised server can propagate malicious content across the network at scale. To mitigate such risks, service providers must integrate security-aware network functions (SNFs) within both hardware and software implementations into their infrastructure. Following this, a fundamental challenge remains: accurately estimating and categorizing security levels (SeLs) within such a framework. This work addresses the above challenge by pioneering a methodology for estimating and classifying the SeL of PFM-hosting servers. First, we introduce the security intensity identifier (SeII), a novel indicator designed to assess security strength. We then propose an innovative SeL calculation methodology that accurately estimates the SeL of PFM-hosting servers. Subsequently, we propose security level indicators (SeLIs) and classify servers with different S-NF sets into different security classes (SeCs) with distinct SeL ranges. Building upon this estimation and classification framework, we design a cost-efficient security complementation scheme for PFMhosting servers, tailored to the diverse security needs of AIGC services. Our extensive experimental results demonstrate that the proposed scheme significantly outperforms state-of-the-art benchmarks, showing an average improvement of 21.77% and 51.77% in implementation cost, respectively