Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

XSS Adversarial Attacks Based on Deep Reinforcement Learning: A Replication and Extension Study

Samuele Pasini, Gianluca Maragliano, Jinhan Kim, Paolo Tonella

TL;DR

The paper tackles the vulnerability of DL-based XSS detectors to adversarial mutations by replicating a state-of-the-art RL-based attack and extending it with an XSS Oracle to test semantic validity. It demonstrates that, when threats to validity are mitigated, adversarial payloads achieve high escape rates (ER) while preserving attack semantics, and shows that preprocessing artifacts largely drive prior high evasion. The authors integrate the Oracle into training to produce robust evaluations, achieving ERs above 96% across detectors and substantially reducing ruin and OOV rates. This work enhances transparency and reproducibility in adversarial XSS research and underscores the need for validation frameworks that preserve payload semantics during evaluation.

Abstract

Cross-site scripting (XSS) poses a significant threat to web application security. While Deep Learning (DL) has shown remarkable success in detecting XSS attacks, it remains vulnerable to adversarial attacks due to the discontinuous nature of its input-output mapping. These adversarial attacks employ mutation-based strategies for different components of XSS attack vectors, allowing adversarial agents to iteratively select mutations to evade detection. Our work replicates a state-of-the-art XSS adversarial attack, highlighting threats to validity in the reference work and extending it toward a more effective evaluation strategy. Moreover, we introduce an XSS Oracle to mitigate these threats. The experimental results show that our approach achieves an escape rate above 96% when the threats to validity of the replicated technique are addressed.

XSS Adversarial Attacks Based on Deep Reinforcement Learning: A Replication and Extension Study

TL;DR

The paper tackles the vulnerability of DL-based XSS detectors to adversarial mutations by replicating a state-of-the-art RL-based attack and extending it with an XSS Oracle to test semantic validity. It demonstrates that, when threats to validity are mitigated, adversarial payloads achieve high escape rates (ER) while preserving attack semantics, and shows that preprocessing artifacts largely drive prior high evasion. The authors integrate the Oracle into training to produce robust evaluations, achieving ERs above 96% across detectors and substantially reducing ruin and OOV rates. This work enhances transparency and reproducibility in adversarial XSS research and underscores the need for validation frameworks that preserve payload semantics during evaluation.

Abstract

Cross-site scripting (XSS) poses a significant threat to web application security. While Deep Learning (DL) has shown remarkable success in detecting XSS attacks, it remains vulnerable to adversarial attacks due to the discontinuous nature of its input-output mapping. These adversarial attacks employ mutation-based strategies for different components of XSS attack vectors, allowing adversarial agents to iteratively select mutations to evade detection. Our work replicates a state-of-the-art XSS adversarial attack, highlighting threats to validity in the reference work and extending it toward a more effective evaluation strategy. Moreover, we introduce an XSS Oracle to mitigate these threats. The experimental results show that our approach achieves an escape rate above 96% when the threats to validity of the replicated technique are addressed.

Paper Structure

This paper contains 26 sections, 4 equations, 3 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Cross-Site Scripting (XSS)
  4. Defending Against XSS Attacks
  5. Adversarial Attacks on XSS Detectors
  6. Reinforcement Learning
  7. Reference Work
  8. Proposed Method
  9. Experiments
  10. Results Reported in the Reference Work
  11. Identified Threats to Validity
  12. Methodology
  13. XSS Oracle
  14. Metrics
  15. Empirical Study
  16. ...and 11 more sections

Figures (3)

  • Figure 1: Method proposed by the reference work (picture taken from the paper by Chen et al. CHEN2022102831)
  • Figure 2: Workflow of the XSS Oracle. A Payload is rendered in a known template of a Web Page and the DOMs of the two pages are compared.
  • Figure 3: Experimental setup: malicious undetected payloads are fed into the Oracle before and after preprocessing