← AI security in plain English

AML.T0093

MITRE ATLAS / ATLAS Techniques

An adversary plants malicious prompts in a victim's public-facing application so that an AI system ingests them later and acts on them. Support tickets, reviews, form submissions, and user profiles are all channels into a system's future context.

Think of it likeWriting instructions in the suggestion box, knowing management reads it aloud at the Monday meeting.

In plain English

Anything a customer can type may end up inside a model's context window later. A support ticket written as an instruction becomes an instruction the moment an AI assistant reads the queue.