Use Redacta safely with AI agents
Guides an AI agent through protecting clinical text with Redacta, releasing restored output to a file, and scanning model output for leaked identifiers. It asks before any release or discard step.
Use it when
- Anonymising a clinical letter before asking an AI to summarise it
- Restoring real identifiers into a file after the AI has worked on tokens
- Checking an AI-written draft for patient identifiers before sharing it
Tools it uses
redactreinstateself_check
Protect this discharge letter before you summarise it, then restore the real details into a file when I'm done.
Protected the letter in session 4f2a: replaced 1 NHS number, 1 patient name and 2 dates of birth with tokens. The self-check passed. The summary is ready, and the scan of it found no leaked values. Restoring the details into a file needs your yes first, and then I'll confirm the destination folder.