Advancing B5G Security: An AI-Augmented Intrusion Detection System using a Real-Time Attack Generator

The complexity and dynamic nature of beyond 5 G (B5G) networks introduce challenges within the realm of cybersecurity, particularly in detecting and mitigating novel threats. This paper presents a comprehensive experimental environment for generating and detecting attacks within 5G and B5G networks. Thus, our setup integrates the Attack Generation Engine (AGE) and an AI-augmented Intrusion Detection System (IDS) deployed on the Centre for Research and Technology Hellas (CERTH) 5G testbed. The attack generator is designed to emulate realistic threat scenarios targeting the Radio Access Network (RAN) and core components, including Denial of Service (DoS) attacks on gNodeBs, Non-Access Stratum (NAS) message tampering and User Plane Function (UPF) exploitation. Using the generated traffic, we construct a labelled dataset representative of realworld 5 G attack patterns. We then evaluate multiple AI-based IDS models, including LSTMs, Graph Neural Networks (GNNs) and Transformer-based architectures, to determine their effectiveness in detecting these threats. The evaluation of our setup will be carried out in the 5 G experimental environment of CERTH, providing critical insights into deploying scalable, intelligent security solutions for next-generation networks.This work builds on attack scenario research initiated in the NATWORK project and represents preliminary research conducted within the early phase of the GuardAI project; further investigations and extended evaluations will be presented in subsequent publications.