Machine learning for 5G security: Architecture, recent advances, and challenges

Abstract The granularization of crucial network functions implementation using software-centric, and virtualized approaches in 5G networks have brought forth unprecedented security challenges in general and privacy concerns. Moreover, these software components’ premature deployment and compromised supply chain put the individual network components at risk and have a ripple effect for the rest of the network. Some of the novel threats to 5G assets include tampering in identity and access management, supply-chain poisoning, masquerade and bot attacks, loop-holes in source codes. Machine learning (ML) in this context can help to provide heavily dynamic and robust security mechanisms for the software-centric architecture of 5G Networks. ML models’ development and implementation also rely on programmable environments; hence, they can play a vital role in designing, modelling, and automating efficient security protocols. This article presents the threat landscape across 5G networks and discusses the feasibility and architecture of different ML-based models to counter these threats. Also, we present the architecture for automated threat intelligence using cooperative and coordinated ML to secure 5G assets and infrastructure. We also present the summary of closely related existing works along with future research challenges.