Audit AI Exposure

Run a post-exposure audit locally. Report counts and trends without printing detected values or placing transcript contents into the current model context.

Boundary

• State that this is an after-the-fact checkup, not a pre-send blocker.
• Scan only transcript files, exports, or paths the user explicitly supplies or clearly places in scope.
• Do not read, grep, summarize, print, or otherwise load original transcript contents into model context.
• Never pass original contents as command-line arguments.
• Never describe a detection occurrence as a confirmed breach or unique leak.
• Keep raw text and detected values on the device. The bundled script writes only aggregate events to its local statistics store.

Scan