Abstract
The transition from 5G to 6G mobile networks necessitates network automationto meet the escalating demands for high data rates, ultra-low latency, andintegrated technology. Recently, Zero-Touch Networks (ZTNs), driven byArtificial Intelligence (AI) and Machine Learning (ML), are designed toautomate the entire lifecycle of network operations with minimal humanintervention, presenting a promising solution for enhancing automation in 5G/6Gnetworks. However, the implementation of ZTNs brings forth the need forautonomous and robust cybersecurity solutions, as ZTNs rely heavily onautomation. AI/ML algorithms are widely used to develop cybersecuritymechanisms, but require substantial specialized expertise and encounter modeldrift issues, posing significant challenges in developing autonomouscybersecurity measures. Therefore, this paper proposes an automated securityframework targeting Physical Layer Authentication (PLA) and Cross-LayerIntrusion Detection Systems (CLIDS) to address security concerns at multipleInternet protocol layers. The proposed framework employs drift-adaptive onlinelearning techniques and a novel enhanced Successive Halving (SH)-basedAutomated ML (AutoML) method to automatically generate optimized ML models fordynamic networking environments. Experimental results illustrate that theproposed framework achieves high performance on the public Radio Frequency (RF)fingerprinting and the Canadian Institute for CICIDS2017 datasets, showcasingits effectiveness in addressing PLA and CLIDS tasks within dynamic and complexnetworking environments. Furthermore, the paper explores open challenges andresearch directions in the 5G/6G cybersecurity domain. This frameworkrepresents a significant advancement towards fully autonomous and secure 6Gnetworks, paving the way for future innovations in network automation andcybersecurity.