AML.T0064

MITRE ATLAS / Gather RAG-Indexed Targets

Adversaries identify the data sources a retrieval augmented generation system indexes, so that later poisoning or manipulation can be aimed at repositories the model will actually read. Public wikis, ticket systems, shared drives, and crawled sites are common targets.

Think of it likeFinding out which newspaper the household reads before planting a story in it.

In plain English

A RAG system is only as trustworthy as the documents it retrieves. Attackers first work out which sources get indexed, because that is where a planted document will be picked up and repeated as fact.