Secure redaction
Short answer: Secure redaction is not a black box drawn on top. It deletes the text and the hidden text layer, covers the pixels in images, rewrites the whole file to remove metadata, attachments, scripts and earlier versions, and then reopens the output to check each item for residue.
What it can find
Selected by default: mainland China ID numbers, Taiwan ID numbers, Hong Kong ID numbers, Thai ID numbers, Singapore NRIC/FIN, bank card numbers, Chinese and Thai mobile numbers, international phone numbers, and email addresses.
Optional: Korean resident registration numbers, Japanese My Number, Indian Aadhaar, US SSN, UK National Insurance numbers, Malaysian ID numbers, and IBAN bank account numbers.
Where a check is possible, matches are validated by check digit or date. For example, Chinese ID numbers use the GB 11643 check code, bank card numbers use the Luhn algorithm, and IBANs use mod 97. This reduces the chance of mistaking an ordinary string of digits for an ID number. You can try the rules in the ID number format tester.
You can also add your own keywords, such as names.
What happens when it runs
- Delete text: characters in the text layer are deleted, along with the hidden OCR text that scans often carry under that area.
- Cover pixels: the matching pixels in the scanned image are covered, and vector graphics that are fully covered are removed.
- Add marks: each area gets a black box, or a box labeled “Redacted”.
- Rewrite the whole file: the file is written again from scratch with no earlier incremental versions kept. Document info and XMP metadata, attachments, JavaScript, forms, annotations, bookmarks, structure tag text, thumbnails, and EXIF, XMP and comments inside JPEG images are removed.
Residue check after saving
After saving, the output file is reopened automatically and checked item by item:
| Check | What is expected |
|---|---|
| Page text and hidden text layer | The original text cannot be extracted or found by search |
| Scanned image pixels | Processed areas are recognized again and must not read as the original text |
| Strings and streams in file objects | The original text does not appear in the file in any encoding |
| Document info and XMP metadata | Empty |
| Annotations, forms, attachments, scripts, bookmarks | Not present |
| EXIF / XMP / comments embedded in images | Removed |
| JBIG2 shared symbol dictionaries | A warning only if present, not a failure |
| Earlier versions | The file has only one version |
Results appear in the save banner and are written to the operation record. For how the checks work, see How to verify that a redaction is clean.
Risk levels do not block redaction
Redaction only removes information and does not rewrite content, so files at any risk level can be redacted.
FAQ
How is this different from replacing with nothing?
Replacing with nothing only removes the visible text. Secure redaction also deletes the hidden text layer, covers pixels in the scanned image, and rewrites the whole file to remove metadata, attachments, scripts, annotations, bookmarks and earlier versions.
Can it find ID numbers in a scanned PDF?
Yes. Both the text layer and scanned images are checked. Scanned pages are fully recognized with the standard recognition model, and then the rules are applied.
Does the results list show the deleted content?
Content in the results list is always partly masked. The operation record (manifest.json) stores only categories and positions, not the deleted text.
Can I add my own keywords?
Yes. For example, names or project code names. Type them in, or import them from the first column of a spreadsheet.