Secure AI/ML-based Control in Intent-based Management System

The emergence of intent-based management (IbM) as a concept for managing and operating telecommunication systems in 6G necessitates the identification of potential threats and risks associated with its adoption and implementation. IbMs leverage automation, artificial intelligence (AI), and machine learning (ML) techniques to manage networks based on high-level and abstract definition of goals and requirements via intents. Although AI/ML models enable the IbM system to dynamically configure network parameters and resources to meet the intents, these models have some intrinsic vulnerabilities against adversarial attacks. In this work, we show the vulnerability of the IbM systems against adversarial attacks which might be originated through a compromised/malicious component in the system. Such attacks can degrade the performance of the IbM system and lead the system to take incorrect decisions. This would result in a propagated effect on the network performance and monetary costs for the operator as well. We further demonstrate the efficiency of the adversarial training as a protection scheme through experiments under various network configurations.