← AI security in plain English

AML.T0006

MITRE ATLAS / Active Scanning

An adversary probes the victim system directly to gather targeting information, in contrast to reconnaissance methods that only read public sources. Scanning can reveal exposed inference endpoints, supported input formats, rate limits, and error behavior.

Think of it likeWalking the perimeter and trying every door handle rather than reading the floor plan from home.

In plain English

Rather than researching from a distance, the attacker pokes the system and watches how it answers. Error messages, timing, and rate-limit responses all leak details about what is behind the endpoint.