← AI security in plain English

AML.T0112

MITRE ATLAS / Machine Compromise

Adversaries compromise a machine by exploiting or manipulating the AI-enabled components running on it, turning a local assistant, agent, or model runtime into the path to full control of the host.

Think of it likeGetting at the whole house through the smart doorbell somebody installed last month.

In plain English

Local AI components are software with system access, and often newer and less hardened than anything else on the machine. Exploiting one is a route to the host itself.