Adversarial Artifacts: Breaking Static ML Malware Classifiers Using Cobalt Strike

No ratings

Presented at 44CON 2025 by

Steve Salinas

As the use of AI for static malware classification continues to grow, it has never been more critical for attackers to understand these systems. This session demystifies how static ML models evaluate binaries and how attackers can systematically break their assumptions. Using Cobalt Strike’s Artifact Kit, we’ll demonstrate how to generate payloads that evade detection by manipulating static features. From injecting misleading byte-level features to reshaping PE metadata, this talk provides a roadmap for red teams and researchers to understand, exploit, and defend against the limits of static AI-powered detection.