A Taxonomy of Adversarial Machine Learning Attacks and Mitigations

This piece critiques NIST’s adversarial machine learning taxonomy, arguing that taxonomies can impose arbitrary structures that quickly become outdated in fast-evolving, adversarial domains. The author questions the taxonomy’s conceptual framing and limited scope, suggests current LLM/ML approaches are short-lived, and briefly contrasts with real-world safety issues in autonomous systems, concluding the framework may not meaningfully advance the field.