Publication lays out “adversarial machine learning (ML)” threats, describing mitigation strategies and their limitations.