[Paper Review] Towards Adversarial Malware Detection: Lessons Learned from PDF-based Attacks

Davide Maiorca, Battista Biggio+1 more|arXiv (Cornell University)|Nov 2, 2018

Advanced Malware Detection TechniquesComputer Science84 references40 citations

TL;DR

A survey of adversarial threats to PDF-based malware detectors, providing taxonomy of PDF malware, learner-based detectors, attack vectors, and defense directions.

ABSTRACT

Malware still constitutes a major threat in the cybersecurity landscape, also due to the widespread use of infection vectors such as documents. These infection vectors hide embedded malicious code to the victim users, facilitating the use of social engineering techniques to infect their machines. Research showed that machine-learning algorithms provide effective detection mechanisms against such threats, but the existence of an arms race in adversarial settings has recently challenged such systems. In this work, we focus on malware embedded in PDF files as a representative case of such an arms race. We start by providing a comprehensive taxonomy of the different approaches used to generate PDF malware, and of the corresponding learning-based detection systems. We then categorize threats specifically targeted against learning-based PDF malware detectors, using a well-established framework in the field of adversarial machine learning. This framework allows us to categorize known vulnerabilities of learning-based PDF malware detectors and to identify novel attacks that may threaten such systems, along with the potential defense mechanisms that can mitigate the impact of such threats. We conclude the paper by discussing how such findings highlight promising research directions towards tackling the more general challenge of designing robust malware detectors in adversarial settings.

Motivation & Objective

• Characterize how PDF files are used as malware infection vectors and why they are challenging for detectors.
• Survey state-of-the-art learning-based PDF malware detectors and their typical architectures.
• Provide a taxonomy of adversarial attacks targeting learning-based PDF detectors and analyze vulnerabilities.
• Identify potential defenses and research directions to improve robustness in adversarial settings.
• Promote security-by-design principles in malware detection systems.

Proposed method

• Provide a comprehensive taxonomy of PDF malware generation methods and corresponding learning-based detectors.
• Describe three-component architecture of ML-based detectors: pre-processing, feature extraction, and classifier.
• Review third-party and custom pre-processing parsers and their capabilities.
• Classify detector features into structural, JavaScript-based, and raw-byte categories and map them to detectors.
• Synthesize known adversarial attack strategies against PDF detectors and discuss practical implementations.
• Outline defense mechanisms and future research directions for robust, adversarially aware malware detectors.

Experimental results

Research questions

• RQ1 What are the main PDF-based malware techniques used in the wild and how have detectors evolved to counter them?
• RQ2 What vulnerabilities exist in learning-based PDF detectors that enable evasion attacks?
• RQ3 How can detector design be improved to withstand adversarial manipulation while maintaining detection performance?
• RQ4 What defense strategies show potential in mitigating adversarial attacks on PDF malware detectors?
• RQ5 What research directions emerge for robust malware detection in adversarial settings?

Key findings

• PDF malware exploits three main channels: JavaScript-based, ActionScript-based, and file embedding, with JavaScript-based attacks historically most common.
• A wide range of detector architectures exists, incorporating static or dynamic pre-processing, various feature types, and classifiers, but all rely on machine learning foundations.
• There is a notable arms race between attackers and defenders, with attackers increasingly exploiting parser vulnerabilities and obfuscation to evade detectors.
• Adversarial attacks can target different components (pre-processing parsers, feature extractors, and classifiers) to evade detection without major code changes.
• Reliance on third-party parsers is common, but introduces security and robustness concerns; however, no parser offers complete coverage of all PDF elements, creating exploitable gaps.
• The work highlights security-by-design as a guiding principle for building more robust malware detectors and outlines directions for future defenses.