Learn how to identify and exploit vulnerabilities in machine learning training infrastructure through this 52-minute conference talk from TROOPERS25. Explore attack vectors targeting ML training systems, understand the security weaknesses inherent in machine learning pipelines, and discover techniques for compromising training environments. Gain insights into the unique security challenges posed by ML infrastructure, examine real-world attack scenarios against training processes, and understand how adversaries can manipulate or disrupt machine learning model development. Master the methodologies for assessing ML training security, learn to identify common misconfigurations and vulnerabilities in training environments, and develop skills for conducting security assessments of machine learning operations infrastructure.