Xai-driven black-box adversarial attacks on network intrusion detectors

Deep Learning (DL) technologies have recently gained significant attention and have been applied to Network Intrusion Detection Systems (NIDS). However, DL is known to be vulnerable to adversarial attacks, which evade detection by introducing perturbations to input data. Meanwhile, eXplainable Artificial Intelligence (XAI) helps us to understand predictions made by DL models and is an essential technology for ensuring accountability. We have already pointed out that XAI is also helpful in identifying important features when making predictions and proposed XAI-driven white-box adversarial attacks on DL-based NIDS. In this study, we extend this work by transitioning from white-box to black-box attacks, thereby increasing the practical applicability of our methods. Furthermore, we implemented our proposed method in a real-world network environment and demonstrated the general effectiveness of our proposed method by targeting multiple NIDS models. Our experimental results show that the proposed black-box attacks achieve high evasion rates without compromising the malicious nature of the attack communications.