MITRE ATLAS · Impact
AML.T0031 · Erode AI Model Integrity
Adversaries may degrade the target model's performance with adversarial data inputs to erode confidence in the system over time. This can lead to the victim organization wasting time and money both attempting to fix the system and performing the tasks it was meant to automate by hand.
Mitigations
- Predictive AI Model Hardening
- Predictive AI Ensembles
- Predictive AI Input Restoration
- Predictive AI Adversarial Input Detection
In the news
No stories mapped to this technique yet.
Mappings are AI-assisted and checked against the framework definitions; corrections welcome.