← AI security in plain English

AML.TA0002

MITRE ATLAS / Reconnaissance

The adversary gathers information about a target's AI systems to plan later operations, using techniques such as Search Open Technical Databases, Search Open AI Vulnerability Analysis, and Active Scanning.

Think of it likeCasing a house from the sidewalk before a break-in: noting which doors have locks, where the cameras point, and when the family leaves for work.

In plain English

Before attacking an AI system, adversaries study it: what model it runs, how it was built, any publicly reported weaknesses, and where its soft spots are.