Why a black box drawn over a PDF is not redaction

Everyone has done this at least once. A PDF needs to be shared, something on it should not be, so you draw a black rectangle over it in Preview, or highlight it in black in Word, or scribble over it with your phone’s markup pen. It looks right on screen. That is precisely the problem: the file looks correct while being completely unredacted, which is why this is the most repeated document-security failure there is.

Why does the text survive a black box?

A PDF is not a picture of a page. It is a structured document. The words, fonts, vector art and images that make up a page live in a content stream. Anything added on top of that page afterwards — a sticky note, a highlight, a stamp, a form field, a drawn shape — lives in a separate list of annotations. The PDF specification, ISO 32000-1, says it in one sentence in §12.5: an annotation “associates an object such as a note, sound, or movie with a location on a page of a PDF document”. With a location. Not instead of what is at that location.

So when you drag a rectangle with a markup or comment tool, you add an annotation. The characters underneath are untouched — same glyphs, same coordinates, same order. The reader paints the page, then paints your rectangle over it. Anything that reads the file rather than looking at it walks straight past the rectangle and finds the text.

That is why every one of these gets it back:

• Select and copy. Drag across the black area, or just select all, and paste into a text editor. Formatting does not survive the paste. Text does.
• Save as text. Any reader’s export-to-text function walks the content stream and writes out what it finds.
• A one-line command. pdftotext file.pdf - prints the lot.
• Delete the annotation. Open the file in a different PDF editor and the rectangle is usually a selectable object you can drag aside.

The UK’s Information Commissioner’s Office describes this failure exactly, in its guidance for organisations disclosing documents to the public: a recipient may reveal the information “simply by copying and pasting a document to a different format (eg pasting a pdf with text covered by simple black rectangles into Notepad)”, because “the information remains in the electronic file if it is only covered by a shape”. Its list of things not to do names the mistake outright.

What it looks like versus what actually happened

The crop row is not hypothetical. In 2023 a bug in Google’s Markup screenshot editor on Pixel phones, tracked as CVE-2023-21036 and fixed in the March 2023 Pixel update bulletin, was described by Google as “a possible failure to truncate images due to a logic error”. The editor opened the existing file for writing, wrote a smaller image into it, and never shortened the file — so the tail of the original stayed on disk past the end of the new one, and a large part of the pre-crop image could be reconstructed. The same class of bug turned up in the Windows Snipping Tool. The lesson is not “distrust Pixel phones”; it is that editing a file in place is a different thing from producing a new file, and only the second reliably destroys data. The thumbnail row is covered in what a photo of your ID gives away besides the ID.

What actually works: removal, not concealment

The principle is old and well settled. NIST SP 800-88 Rev. 1, the US federal guideline on media sanitisation, defines sanitisation as a process that renders access to the target data “infeasible for a given level of effort”. Making data invisible is not on that scale. Making it absent is.

There are three honest ways to get there.

1. Use a real redaction tool, and know which one it is. In Acrobat Pro, the Redact tool is not the same thing as the shape or highlight tools sitting next to it. Redaction is a two-step operation, and the PDF specification defines it that way: first you mark regions, then you apply the redaction, at which point the content inside those regions is removed from the document and a mark is drawn in its place. The specification is unusually blunt about the requirement — a conforming reader “shall remove all traces of the specified content”, and if part of an image falls inside a redaction region “that portion of the image data shall be destroyed; clipping or image masks shall not be used to hide that data.” If you have Acrobat Pro, this is your fix, and it is the single most useful distinction in this guide.

2. Flatten the page to pixels. Export or print the page to an image format with no layers and no text, then, if you need a PDF back, rebuild it from that image. The ICO calls this “roundtrip redaction” and recommends it for exactly this reason. Once the page is pixels there is no content stream to extract and no annotation to delete. The cost is that the document stops being searchable and machine-readable — for a copy of an identity document, not a cost at all.

3. Print it, redact the paper, scan the paper. Unglamorous and completely effective, because the scan is generated from what the sensor saw. Germany’s Lower Saxony data protection authority recommends precisely this when you have no software that removes content without residue, and warns that a marker pen alone often fails once the sheet is held up to a light or scanned with filters. Cover the area with opaque tape or cut it out, then copy.

What does not work, plainly: blurring, pixelating, shrinking an image so the text is “too small to read”, matching text colour to the background, or covering something and hoping. The reversibility of blur and pixelation is covered in blurring your ID yourself vs using a dedicated app.

Verify before you send

Whatever method you used, check it. None of this needs any technical skill and all four together take about two minutes:

1. Select all, copy, paste into a text editor. If the covered words show up, you are not done.
2. Export the file as text and read the result. This catches what a drag-select misses.
3. Open the file in a second app. If the black box is draggable there, it is draggable for the recipient.
4. Look at the file size. Real redaction removes bytes; an annotation only adds them.

Run these on the file you are about to send, not on the one you edited — those are sometimes different files, particularly if a cloud editor made a copy along the way.

Where this sits in the bigger picture

If a black-boxed PDF has already gone out, the recovery steps are the same as for any premature disclosure: see you already sent an unredacted ID copy. If the document is an identity document, which fields to cover is in how to redact a passport or ID card — and before you redact anything, check that redaction is even the right answer, because for banks, statutory right-to-work checks and government procedures it is not: when redacting your ID will not work.

One honest note on tools. Anonymize my ID works on a photograph of an identity document and exports a flattened image, so the covered pixels do not exist in the output file at all. That makes it the right tool for “I need to send a covered copy of my passport or ID card”. It is not a PDF redaction tool: if what you have is a multi-page PDF — a contract, a bank statement, a court filing — use a proper redaction tool, flatten it to an image, or redact it on paper, then run the four checks above.

Check whether a PDF is really redacted

1. Select everything and paste it into a text editor. Open the PDF in any reader, press Ctrl+A (Cmd+A on a Mac) then Ctrl+C, and paste into Notepad, TextEdit in plain-text mode, or any code editor. If the words you covered appear in the paste, the file is not redacted. Stop and fix it before sending.
2. Export the file as plain text. Most readers offer File → Export To → Text, or Save As → Text. This walks the whole document rather than only what your selection happened to catch, so it finds text a drag-select can miss — including content sitting outside the visible page area.
3. Open the file in a second application. Open the same PDF in a different reader or editor — a browser, a free PDF editor, anything that is not the tool you drew in. If the black box can be clicked, dragged or deleted, it is an annotation, and your recipient can remove it just as easily.
4. Compare the file size against the original. Redacting removes content, so a genuinely redacted file is usually smaller, and a rasterised one is usually very different in size. If the covered file is the same size as the original or slightly larger, nothing was taken out — an annotation only adds bytes.
5. If you have a terminal, extract the text directly. Run pdftotext redacted.pdf - to print everything in the content stream to your screen. It takes one line and it is the same thing any recipient could do. Nothing sensitive should appear in the output.

Frequently asked questions