AI SECURITY CASES × FRAMEWORKS
NIST Adversarial ML (2025) · Categories and AI Security Cases
Explore AI security cases, sources, and defensive analysis across 17 populated NIST Adversarial ML (2025) categories.
A taxonomy and terminology of attacks and mitigations for AI systems, published by NIST (March 2025).
- NISTAML.011 · Model Poisoning (Availability) 0 cases
- NISTAML.012 · Clean-label Poisoning 2 cases
- NISTAML.013 · Data Poisoning 5 cases
- NISTAML.014 · Energy-latency 0 cases
- NISTAML.015 · Indirect Prompt Injection 66 cases
- NISTAML.018 · Prompt Injection 26 cases
- NISTAML.021 · Clean-label Backdoor 0 cases
- NISTAML.022 · Evasion 16 cases
- NISTAML.023 · Backdoor Poisoning 2 cases
- NISTAML.024 · Targeted Poisoning 3 cases
- NISTAML.025 · Black-box Evasion 4 cases
- NISTAML.026 · Model Poisoning (Integrity) 6 cases
- NISTAML.027 · Misaligned Outputs 51 cases
- NISTAML.031 · Model Extraction 9 cases
- NISTAML.032 · Reconstruction 0 cases
- NISTAML.033 · Membership Inference 0 cases
- NISTAML.034 · Property Inference 0 cases
- NISTAML.035 · Prompt Extraction 3 cases
- NISTAML.036 · Leaking Information from User Interactions 3 cases
- NISTAML.037 · Training Data Attacks 2 cases
- NISTAML.038 · Data Extraction 34 cases
- NISTAML.039 · Compromising Connected Resources 31 cases
- NISTAML.051 · Model Poisoning (Supply Chain) 4 cases
An empty category does not establish safety. A category relation does not prove a defense is deployed or effective.