← AI security in plain English

AML.T0018

MITRE ATLAS / Manipulate AI Model

Adversaries alter a model artifact or its bundled components to change system behavior, introduce malicious code, or install persistent malicious functionality. The change may target weights, architecture, or the serialization format itself.

Think of it likeRewiring an appliance so it works normally until the one day you flip a hidden switch.

In plain English

A model file is code and data together. Editing it can install behavior that survives restarts and redeployments, because the backdoor ships inside the artifact everyone trusts.