AML.T0001

MITRE ATLAS / Search Open AI Vulnerability Analysis

Adversaries look for pre-existing published research on weaknesses in the class of model a target uses, including known attack methods, published adversarial examples, and reproduction code. Reusing someone else's analysis is cheaper than developing an attack from scratch.

Think of it likeChecking whether a lock model already has a known bypass posted online before trying to pick it yourself.

In plain English

Most AI attacks do not need to be invented. If the target runs a well-known model family, the weaknesses are usually documented in public research, often with working code attached.