Reading an ID with AI: upload, review, apply — and why it never writes anything without you

1. Upload the ID photo or scan to Client › Documents. JPG, PNG or PDF. It is encrypted on arrival and opened only by roles that may view client documents.
2. Read with AI. The image is sent to the reader your workspace is configured with (set in Settings › Automation with your own key; nothing is sent if no key is set). The reader returns text; the platform maps it to fields.
3. Review. Each field is a row: proposed value, tick box, and a note if it differs from what is already on record. Fields the reader was unsure of arrive unticked.
4. Tick and edit. Anything you untick is discarded. Anything you edit is your value, not the reader's.
5. Apply. Only ticked rows are written. The audit log records which fields were applied from a reading, by whom, when.

What is sent, what is kept

Sent: the image, once, to the reader you chose. Kept by the platform: the image (sealed), the fields you applied, and the audit entry. Not kept: the reader's raw response, once you have applied or discarded it. Whether the reader retains anything is governed by your agreement with that provider; the settings page links to it.

Why it cannot auto-write

Because an ID number written wrong silently is worse than one never written, and because under RA 10173 you are the one accountable for the accuracy of what you hold. The tick boxes are not a convenience feature; they are the control that makes using a reader defensible.