Confused Learning: Supply Chain Attacks Through Machine Learning Models

Explore the hidden vulnerabilities in machine learning models and their potential for supply chain attacks in this eye-opening Black Hat conference talk. Delve into the surprising world of ML security, where models can harbor malware while maintaining accuracy, and learn how attackers can exploit the trust placed in ML repositories. Discover novel techniques for distributing malware through ML models, compromising major companies, and gaining access to sensitive data environments. Gain insights into writing ML malware, its distribution methods, and post-compromise looting techniques. Examine available tools and techniques for analyzing potentially malicious models, and explore threat hunting strategies for detecting machine learning malware in the wild. Benefit from the speakers' expertise as they share open-source code, practical advice on mitigation and prevention, and valuable lessons learned from their research and real-world experiences.