Data Security

Can a Redacted PDF Be Unredacted? 9 Ways Redaction Fails

Only if it failed. Black boxes that copy-paste, hidden OCR text, bookmarks, metadata: 9 ways redaction fails, real cases and a 5-minute test for your file.

Can a Redacted PDF Be Unredacted? 9 Ways Redaction Fails

Yes, but only if the redaction failed. A redaction that worked deletes the text from the file, so the words themselves can’t be recovered (at most, the size and position of the box give clues; see failure 9). A failed one only hides it: the words are still there under a black box, in a hidden text layer, in a bookmark, in an attachment or in an earlier version of the file. Anyone who opens the PDF can find them. Below are the nine ways this happens, three real cases, and a five-minute test you can run on your own file before you send it.

Real redaction vs. a black box

A PDF is not a picture of a page. It is a set of layers and containers: the page’s text and drawings, images, annotations and comments, bookmarks, form fields, attachments, metadata, and sometimes earlier versions of all of these. What you see on screen is only what the viewer draws.

  • A black box is one more thing drawn on top. The text underneath is unchanged. Select it, search for it, or delete the box, and it’s back.
  • A real redaction removes the characters (or image pixels) from the page and puts a box where they were. There is nothing under the box.

A PDF drawn as a stack of layers: the black box sits on the top layer, the tenant's name is still on the page-text layer below it, and under those are comments and form fields, bookmarks and attachments, and metadata and earlier saves

“Can this PDF be unredacted?” comes down to which of the two it is, and to whether the same information also sits in one of the other containers.

Test your own PDF in five minutes

Run these checks on the file you are about to send, in the viewer you already have. Pick one or two words you redacted to test with.

  1. Search. Press Ctrl+F (Cmd+F on a Mac) and search for a redacted word. If the viewer finds it, the redaction failed.
  2. Select all and paste. Press Ctrl+A, then Ctrl+C, and paste into a plain text editor (Notepad, TextEdit in plain-text mode). Read what comes out where the boxes were. Text there means it was covered, not removed.
  3. Open it in a second viewer. Open the file in your browser (Chrome, Edge and Firefox all show PDFs). If a box disappears, turns into an outline or shows text, it was an annotation or a redaction mark that was never applied.
  4. Open the side panels. Look at Bookmarks, Attachments, Comments and Layers. Each can hold the text you removed from the page.
  5. Check the properties. Open File → Properties (Document Properties). Read the title, author, subject and keywords fields.

If a page is a scan, search for a word you can see elsewhere on that page. If the search finds it, the page has a hidden OCR text layer, and step 1 and 2 matter even more.

A file that passes: the search finds nothing, the paste has gaps (or your placeholder labels) where the boxes are, the panels are empty or harmless, and the properties don’t name anyone.

For technical readers: command-line checks

These use the free poppler-utils tools and ExifTool.

CheckCommandWhat to look for
Text under the boxespdftotext -layout file.pdf -The redacted words anywhere in the output
Document metadatapdfinfo file.pdf and pdfinfo -meta file.pdfNames, case numbers or file paths in Title, Author, Subject, Keywords or the XMP packet
Image and other metadataexiftool -a -G1 -ee file.pdfCaptions, names, software history
Attachmentspdfdetach -list file.pdfAny attached file you didn’t mean to send
Imagespdfimages -list file.pdfImages you don’t expect, such as a full-page scan behind the text
Earlier versionsgrep -a -o '%%EOF' file.pdf | wc -l1 is normal. 2 can mean “fast web view” or one incremental save. More means the file holds earlier versions

The 9 ways PDF redaction fails

1. Covering the text instead of removing it

The most common failure: a black rectangle, a black highlight or a black text background drawn over the words. The page looks redacted; the text layer is untouched. Copy-paste, search or text extraction reads it.

Two copies of the same line. Covered: copy-paste still gives the tenant's name. Removed: the paste gives only the box's label, P1, where the name was

This is how the redactions failed in a court filing by Paul Manafort’s lawyers in January 2019. The PDF showed black bars, but reporters could select the blacked-out lines and paste them into another document, which revealed that prosecutors accused Manafort of lying about sharing 2016 campaign polling data with Konstantin Kilimnik (BuzzFeed News, Axios). In December 2025, readers of the US Justice Department’s release of Epstein case files found the same flaw in at least one court document: the blacked-out paragraphs could be highlighted and pasted into a word processor (The Guardian via Yahoo News, Snopes).

“Flattening” the file afterwards doesn’t fix it. Flattening merges the boxes into the page content, so the box can no longer be moved, but the text drawn underneath is still in the same content and still copies. The fix is to remove the characters themselves.

Check: steps 1 and 2 of the test.

2. Hidden text layers: OCR, white text and form fields

Scanned PDFs often carry an invisible OCR text layer so you can search them. If you black out the words on the scanned image, the image is covered, but the OCR text behind it still holds the words. A redaction has to remove the text in both places.

Other hidden text: text in the same colour as the background, text placed off the visible page, and form fields whose value is still stored after the field was hidden or covered.

Check: search for a redacted word on the scanned page (step 1), and paste the text out (step 2).

3. Redaction marks that were never applied, and comments

Tools like Adobe Acrobat redact in two steps: you mark the areas, then you apply the marks, and only applying removes the content (Adobe). A file saved after marking but before applying still has all the text under the marks. Some viewers draw the marks as black boxes, others show only an outline or nothing, which is why “my redaction shows text in a different reader” is a common question.

A reviewer's comment next to a redacted name in a letter reads: redact the client's name here, Emma Walker

Comments and sticky notes are a second leak. A reviewer’s note such as “redact the client’s name here” may contain the name. Comments live outside the page text, so removing the text doesn’t remove them.

Check: step 3 (second viewer) and the Comments panel in step 4.

4. Document metadata

A PDF stores data about itself: title, author, subject, keywords, the software that made it, and creation and modification dates. This is often filled in automatically from the source document. A title copied from the first line of a memo, or a file name with a case number and a client’s surname, survives a perfect redaction of the pages. Any viewer shows it under Properties, and search engines index it.

Check: step 5, or pdfinfo and exiftool -ee.

5. Bookmarks and links

Bookmarks (the outline in the sidebar) are a separate list of text. They are often generated from the headings, so they repeat what the headings said before you redacted them.

A PDF viewer: the page heading is redacted, but the Bookmarks panel still shows it: Payment, one third upfront

In January 2021 the European Commission published its redacted vaccine supply contract with AstraZeneca. The pages were redacted, but the PDF’s bookmarks were not: anyone who opened the Bookmarks tab could read redacted passages, including the payment terms (a third of €336 million upfront within 30 days of signing). The price per dose stayed hidden. The Commission replaced the file with one that had no bookmarks (The Register).

Links have the same problem. A link has visible text and a hidden target. Redact the visible text, and the target (a URL with a name or account number, a file path with a user name) is still in the file.

Check: the Bookmarks panel in step 4; click or hover over links near redacted text.

6. Attachments and image metadata

A PDF can carry other files: a spreadsheet with the data behind a table, the original Word document, an email. Redact the table on the page, and the attached spreadsheet still has every row.

A salary table with every name and amount redacted, and the attached salaries.xlsx that lists them all in full

Images carry their own metadata (EXIF or XMP), such as a caption, the photographer’s note or the name of the person in the photo. Blurring a face does nothing to the caption stored in the image. Attachments have to be deleted on their own.

Check: the Attachments panel in step 4, pdfdetach -list and exiftool -ee.

7. Incremental saves

The PDF format lets an editor save changes by appending them to the end of the file instead of rewriting it (ISO 32000-1, section 7.5.6). The earlier version stays in the file, unused but readable. Delete a paragraph, add a box, save incrementally, and the old paragraph is still in the bytes.

Acrobat Pro’s Sanitize Document removes “hidden data from previous document saves”, along with metadata, attachments, bookmarks and comments (NSA guide to redaction in Acrobat X). Without it, the safe route is to redact with a tool that writes a new file and to send that file.

Check: the %%EOF count in the command table.

8. Missed occurrences

A redaction is only as good as the occurrence you missed. The name you removed on page 2 can still be in a running header, a footer, a caption, a footnote, a table of contents, an index or table of authorities, or a file name printed at the bottom of each page. Search-based redaction misses spellings: “M. Hansen” when you searched for “Mette Hansen”, or a name split across two lines.

Check: search for every variant of each name or number, not only the one you redacted first.

9. The box itself gives clues

Even when the text is gone, the box can leak. Its width tells you roughly how long the word was, and with a proportional font and a list of candidate names, that narrows the options. Researchers went further: they showed that sub-pixel shifts in the positions of the characters that remain around a redaction can be used to recover first and last names. They tested 11 popular redaction tools, including Adobe Acrobat, found that they leak information about the redacted text, and deredacted hundreds of real-world redactions, including some in FOIA responses (Bland, Iyer and Levchenko, PoPETs 2023).

It is a lower risk than failures 1 to 8, which need little or no skill, but it is a real one: the attack works even when the text under the box was removed. A fixed-width label on its own doesn’t help, because the characters around the box keep their positions. The paper’s stronger fixes: redact the whole line (or the surrounding words) rather than a box cut to the name, or replace the name with meaningless text in the source document before you make the PDF.

The 9 failures at a glance

FailureWhere the text survivesHow to check
1. Covering, not removingPage text under the boxSearch; select all and paste
2. Hidden text layersOCR layer, white text, form fieldsSearch a scanned page; paste
3. Unapplied marks, commentsPage text; comment textSecond viewer; Comments panel
4. MetadataTitle, author, subject, keywords, XMPFile → Properties; pdfinfo, exiftool
5. Bookmarks and linksBookmark titles; link targetsBookmarks panel; hover links
6. Attachments, image metadataAttached files; EXIF/XMPAttachments panel; pdfdetach, exiftool
7. Incremental savesEarlier versions in the same fileCount %%EOF
8. Missed occurrencesHeaders, footers, indexes, other spellingsSearch every variant
9. Box size and glyph positionsWidth of the box; position of surrounding charactersRedact whole lines, or replace the name before making the PDF

A failed redaction can be a data breach

If the text you meant to remove is personal data, a PDF sent with a failed redaction can be a personal data breach under the GDPR: “a breach of security leading to the accidental or unlawful destruction, loss, alteration, unauthorised disclosure of, or access to, personal data transmitted, stored or otherwise processed” (GDPR Art. 4(12)). The controller then has to notify the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware of it, unless the breach is unlikely to result in a risk to people’s rights and freedoms (Art. 33(1)). Our guide to GDPR-compliant redaction covers what to remove and when.

Questions people ask

What is a redaction error? Information that was meant to be removed from a document but can still be read or recovered from the released file. The causes are the nine above. The term is also used for the opposite mistake: removing information that should have stayed.

Why does my redaction show text when I open the PDF in a different reader? Because the box is an annotation or an unapplied redaction mark, not a redaction. Viewers draw annotations differently, and the text was never removed. Apply the redaction (or redact again with a tool that removes the text) and save a new file.

Can a properly redacted PDF be recovered? Not from the file. Once the characters are deleted and the file is rewritten without earlier versions, the text isn’t in it. What can still leak is everything outside the page text (failures 4, 5, 6 and 8) and the clues in failure 9.

Redact it so it holds

redactpdf.io removes the characters and image pixels under each box rather than covering them, and it clears the document properties (title, author, subject, keywords) and the XMP metadata, including on images. It is free for a single document of up to 25 pages and 5 MB, with no account. The AI suggests what to redact; you check the suggestions, add anything it missed, and download the result.

What it doesn’t remove: attachments, bookmarks, JavaScript, layers, and comments or form values outside the redacted areas. Those are failures 2, 3, 5 and 6 above. Delete them before you upload, or strip them from the file afterwards.

Then run the five-minute test on the downloaded file, whatever tool you used. No tool catches every edge case, ours included: a box drawn thinner than the line of text, for example, can leave characters behind, so draw boxes over the full height of the text. If you upload documents to any online service, here is what to check first. For more than one document at a time, see pricing. Organisations with folders of documents to redact can use Redact Folders on ante.dk.

Try it now

Looking for a Redaction Solution?

Try redactpdf.io – our free and dedicated redaction tool that makes redacting documents secure and effortless. Built-in AI features automatically identify personal data for quick and reliable redaction.