ai-redact.com

The model scans for personal identifiers as it reads the page: full names, email addresses, phone numbers, Social Security numbers, account and routing numbers, and health information. It works across common business document types rather than one narrow form. PII detection runs page by page, and you see each hit before anything is removed, so a false positive costs one click.

Redacted characters are taken out of the PDF content stream, not covered with a shape. Copy the page into a text editor afterward and the data isn't there. That distinction matters for anything heading to a court, a regulator, or a journalist, where a visual overlay has failed people before.

You can push several files at once instead of one at a time. Each one runs as its own job with its own status, and clearing a job from the list deletes the redacted file from the servers. For a disclosure set or a month of statements, that saves the babysitting.

Scanned pages and photographed documents go through optical character recognition first, so you can redact scanned PDF files that have no text layer at all. The same detector then runs on the extracted text. Without OCR, a scan would sail through untouched because there's nothing for a text parser to read.

Every match lands in a list you can edit. Confirm the obvious ones, drop the ones the model misread, add anything it missed. This is the step that makes automated document redaction usable in a compliance workflow, since a person signs off on the final file.

There's no installer and no desktop app. Uploads travel encrypted, and the site says files are never stored after processing. Removing a job deletes the output permanently, so the redacted copy doesn't sit on someone else's disk.