attackGAN: Adversarial Attack against Black-box IDS using Generative Adversarial Networks
Shuang Zhao, Jing Li, Jianmin Wang, Zhao Zhang, Lin Zhu, Yong Zhang
Abstract
With the rapid development of Internet of Things technology, a large number of devices are connected to the Internet of Things, and at the same time, a large number of network attacks and security threats are introduced. Intrusion detection system (IDS) is one of the effective methods for protecting network. With the rise of artificial intelligence technology, intrusion detection system based on ML/DL is widely applied. However, neural network is vulnerable to adversarial perturbation. Most of existing adversarial attacks cannot guarantee the basic function of traffic data. In this paper, we propose an improved adversarial attack model based on Generated Adversarial Network called attackGAN, and design a new loss function to achieve effective attack against the black-box intrusion detection system on the premise of ensuring network traffic functionality. Experiments show that the proposed attackGAN can improve the success rate of adversarial attack against the black-box IDS compared with Fast Gradient Sign Method (FGSM), Project Gradient Descent (PGD), CW attack (CW) and the GAN-based algorithms.