← AI security in plain English

AML.T0081

MITRE ATLAS / Modify AI Agent Configuration

Adversaries modify the configuration files that govern AI agents on a system, changing tool permissions, system prompts, model endpoints, or safety settings.

Think of it likeQuietly rewriting the standing orders pinned above the night guard's desk.

In plain English

Agent behavior lives in config: which tools it may call, what its system prompt says, which endpoint it talks to. Edit that file and you have changed the agent permanently without touching its code.