← AI security in plain English

AML.T0049

MITRE ATLAS / Exploit Public-Facing Application

Adversaries exploit a weakness in an internet-facing computer or program using software, data, or commands to cause unintended behavior. In AI systems this often means the serving layer, the notebook environment, or the orchestration around the model rather than the model itself.

Think of it likePrying open a badly fitted window on the ground floor instead of arguing with the doorman.

In plain English

The model is often the best-defended part of the stack. The inference server, the notebook host, and the API gateway around it are ordinary software with ordinary vulnerabilities.