The Practical Divide between Adversarial ML Research and Security Practice - A Red Team Perspective

Explore a 21-minute conference talk from USENIX Enigma 2021 that delves into the practical divide between adversarial machine learning research and security practices from a red team perspective. Gain insights from Hyrum Anderson of Microsoft as he discusses the significant gaps between academic advancements and industry needs in ML security. Learn about sobering lessons from a Machine Learning Red Team engagement at Microsoft, including the importance of traditional security measures and the low awareness of ML vulnerabilities outside of security applications. Discover why most organizations struggle to protect their ML models despite extensive research in the field, and understand the challenges in translating academic tools and techniques to business needs. Examine real-world examples, red team attacks, and lessons learned to better grasp the current state of ML security and its implications for corporations and government entities.