AI/ML specific Threat Modeling in Mobile Networks

Abstract

The incorporation of AI and ML into RAN is enhancing mobile networks with better optimizations and automation. However, using AI/ML introduces new security vulnerabilities unique to them and cannot be addressed by traditional security evaluation frameworks. This thesis addresses the gap by proposing a tailored threat modeling framework for security evaluation of ML based features in RAN. The threat modeling framework is built upon established guidelines from OWASP, STRIDE and LINDDUN and incorporates an attack library based on NIST but redefined in the context of RAN.

This framework was designed to evaluate features based on predictive ML algorithms and deployed in a physical base station environment. It provides a structured approach to identify, categorize and mitigate the security risks. Moreover, the framework’s design allows for future expansion to include generative ML algorithms and cloud based RAN deployments.

This research fills a critical gap in the literature by extending the AI/ML security evaluation to the unique requirements of RAN, contributing as a valuable resource for security evaluation of AI/ML integration in the next generation mobile networks.