AML.T0108

MITRE ATLAS / AI Agent

Adversaries abuse an AI agent already present on the victim system as their command and control mechanism, issuing instructions the agent carries out on their behalf.

Think of it likeUsing the household's own assistant to relay your orders, so no stranger is ever seen on the property.

In plain English

An agent that already runs, already has credentials, and already talks to the outside makes an ideal control channel. Nothing new needs installing for the attacker to keep issuing commands.