Adversarial Attacks on Pre-trained Deep Learning Models for Encrypted Traffic Analysis

For web security, it's essential to accurately classify traffic across variousweb applications to detect malicious activities lurking within network traffic.However, the encryption protocols for privacy protection, such as TLS 1.3and IPSec, make it difficult to apply traditional traffic classification methodslike deep packet inspection (DPI). Recently, the advent of deep learninghas significantly advanced the field of encrypted traffic analysis (ETA),outperforming traditional traffic analysis approaches. Notably, pre-traineddeep learning based ETA models have demonstrated superior analyticalcapabilities. However, the security aspects of these deep learning modelsare often overlooked during the design and development process. In thispaper, we conducted adversarial attacks to evaluate the security of pre-trainedETA models. We targeted ET-BERT, a state-of-the-art model demonstratingsuperior performance, to generate adversarial traffic examples. To carry outthe adversarial example generation, we drew inspiration from adversarial attacks on discrete data, such as natural language, defining fluency from anetwork traffic perspective and proposing a new attack algorithm that canpreserve this fluency. Finally, in our experiments, we showed our target modelis vulnerable to the proposed adversarial attacks.