← AI security in plain English

AML.T0099

MITRE ATLAS / AI Agent Tool Data Poisoning

Adversaries manipulate data in a victim-controlled, trusted, or connected source that an agent reaches through one of its tools, so the agent acts on falsified information.

Think of it likeChanging the numbers in the ledger the accountant checks, rather than arguing with the accountant.

In plain English

Agents trust what their tools return. Corrupting the database, ticket queue, or file share behind a tool means the agent reasons correctly from data that is a lie.