NIST AI 100-2 E2025 - Adversarial Machine Learning: A Taxonomy and Terminology of Attacks & Mitigations
- Publisher: NIST
- Status:
final - Version:
2025 - Release Date:
2025-03-24 - Date Added:
2025-10-30 - Source URL: https://csrc.nist.gov/pubs/ai/100/2/e2025/final
Summary
This report provides a taxonomy of concepts and definitions of terminology within the field of adversarial machine learning (AML). It is meant to provide common language, informing future standards and guides for assessing and managing AI system security.
Key Takeaways
- Taxonomies cover multiple dimensions of adversarial risk, including system type, lifecycle stage, and attacker capabilities.
- Distinguishes between predictive and generative AI, and the vulnerabilities that are inherent to each type
- Provides mitigation techniques tied to each threat
Additional Sources
Tags
taxonomy, data-poisoning, evasion, mitigations
License
Public-domain