One page per term. Each short enough to quote.

1. 01

   AI DLP

   data loss prevention for the prompt layer

2. 02

   Shadow AI

   AI tool use outside sanctioned channels

3. 03

   On-device redaction

   masking at the point of detection, before anything is recorded

4. 04

   Sanctioned AI tool

   the contractual distinction that should drive severity

5. 05

   Coverage gap

   the surfaces a control does not see

Longer questions — is this legal, does ChatGPT train on it, how do I stop it — are answered in Answers.