Before your first enterprise deal

Use this stage when a buyer asks how your AI handles their data and what happens when it follows the wrong instruction. Start with a page your own team can read. Name what you hold, where outside text enters, what the model can do and what a person must approve.

Keep each list concrete. Include the tools the feature can call, the data those tools can read and the channels they can send on. Mark any feature that combines private data, untrusted content and a way to send data out. That combination needs a boundary you can point to in code and test.

Ask another person to read the page without a walkthrough. They should be able to name the worst outcome in one sentence and identify the approval step that limits it. If they cannot, revise the description before using it in a buyer's review.

Done means the page describes the feature you run today. Link the controls and evidence you have, and leave claims you cannot support out of your answer.

Know the limit

A threat model is a description of boundaries and failure paths. It is not a third-party test report or evidence that every boundary works.