Open the redactor →

Redactron / Guides

The Redaction Hall of Shame: nine real redaction failures

Redactron team · Updated 30 September 2026

A black rectangle drawn over a paragraph is not a hole in the document. It is a shape sitting above the text, in a layer of its own. The words underneath are still there, in the same order, at the same coordinates. Anyone who opens the file with a tool that reads the text layer instead of the picture gets the original sentence back.

Covering and removing are two different operations. Almost every redaction failure below is the same mistake: someone did the first and believed they had done the second.

The nine redaction failures below are verified and sourced. They span twenty years and involve US and UK public bodies, two of the world's largest companies, a newspaper of record and a law firm. The mechanism barely changes.

The nine cases at a glance

YearWhoWhat escapedHow
2005US ArmyClassified detail, Calipari reportCopy and paste
2009TSA contractorAirport screening manualCopy and paste
2012Quinn EmanuelApple's patent license termsIncomplete redaction
2014The New York TimesAn NSA employee's nameCopy and paste
2019Manafort defence teamContacts with a Russian associateCopy and paste
2020Federal court filingNames in a sealed depositionAlphabetised index
2023Police Service of Northern IrelandDetails of 9,483 officersHidden worksheet
2025MetaApple, Google and Snap materialRemovable black boxes
2025US Justice DepartmentVictims' names; text under black boxesNames never redacted; copy and paste

2005: the US Army's Calipari report

In March 2005, US soldiers at a Baghdad checkpoint opened fire on a car carrying the freed Italian journalist Giuliana Sgrena. The Italian intelligence officer Nicola Calipari was killed. The Army investigation came out at the end of April with heavy black bars over names, troop positions and attack statistics.

The bars were cosmetic. Within days, readers copied the pages into a word processor and read the report in full, including attack counts for Baghdad and internal recommendations on rules of engagement. The Register published the recovered material on 3 May 2005. The Pentagon pulled the file, but copies had spread. The NSA later issued written guidance on redacting Word and PDF documents.

Lesson: this failure mode was documented, at the highest level of government, twenty years ago. Every case below repeats it.

2009: the TSA screening manual

In December 2009 a 93-page TSA Standard Operating Procedures manual turned up on the Federal Business Opportunities website, posted by a private contractor during a procurement. The sensitive passages had been covered, not deleted, so the text under the boxes came straight out.

It included sample CIA, congressional and law enforcement credentials, and the proportion of checked bags hand-searched for explosives. Senator Susan Collins said the manual "provides a road map to those who would do us harm." ABC News reported the breach on 9 December 2009. DHS began personnel action against contractor staff and TSA supervisors, and its Inspector General opened a review.

Lesson: a contractor's redaction is your disclosure. The same goes for any outside tool that touches the file, which is why it is worth asking whether it is safe to redact a PDF online.

2012: a law firm's missed redaction in Apple v. Samsung

In March 2012 an associate at Quinn Emanuel prepared an expert report for disclosure to the firm's client, Samsung. It carried the confidential terms of Apple's patent licenses with Nokia and other manufacturers, marked for attorneys' eyes only. The associate redacted the report incompletely and missed those terms. The file went to an FTP site Samsung staff could reach, and more than 200 people who should not have seen those terms did.

In January 2014, Magistrate Judge Paul Grewal ordered Quinn Emanuel to reimburse Apple, Nokia and their outside counsel for the cost of litigating the matter. Nokia alone claimed more than $1.1m in fees. Bloomberg Law has the full account.

Lesson: the job is not finished when someone marks the text. It is finished when someone opens the exported file and checks.

2014: The New York Times and an NSA document

In January 2014 the Times published a leaked NSA presentation as a PDF alongside its reporting. Black boxes covered details meant to stay hidden, including the name of the NSA employee who prepared the document and the name of a targeted network.

The boxes sat on top of live text. Cryptome noticed, and showed that three of the redactions could be read by highlighting the text, copying it and pasting it. The Times called it a production error, removed the file, redacted it again and republished it. Techdirt reported it on 28 January 2014.

Lesson: a newsroom that handles leaked documents every week still shipped live text under black boxes. Test the file you publish, not the page you see.

2019: the Paul Manafort filing

On 8 January 2019, lawyers for Paul Manafort answered the Special Counsel's claim that their client had lied to investigators. Passages in the brief appeared blacked out. They were not. The rectangles were annotations over live text in an unflattened PDF.

Within minutes of the filing hitting the docket, reporters could copy and paste redacted text straight out of it. The recovered passages described Manafort's contacts with Konstantin Kilimnik, including the sharing of polling data, which was exactly what the defence had fought to keep sealed. Vice covered it the next day.

Lesson: a court filing is public the instant it is docketed. There is no window in which to notice and fix.

2020: the Ghislaine Maxwell deposition

This one is different, and it is why the copy-paste test is necessary rather than sufficient. In October 2020 a federal court unsealed Ghislaine Maxwell's 2016 deposition from Virginia Giuffre's defamation suit. The redactions held: the covered words really were gone from the text layer.

The transcript shipped with an alphabetised index of every word in it, redacted in place. That told readers where each hidden word fell in the alphabet, how often it appeared and on which pages. A Slate team reconstructed the covered names from that structure on 22 October 2020. No tooling flaw was involved.

Lesson: what surrounds a redaction can rebuild it. Indexes, word counts and box widths are all data about the thing you removed.

2023: the Police Service of Northern Ireland spreadsheet

On 8 August 2023 the PSNI answered a freedom of information request, submitted through the WhatDoTheyKnow website, with an Excel file. Staff had built the published answer from a larger internal download and deleted the worksheets they could see. One worksheet stayed in the file, hidden, holding the surname, initials, rank and role of all 9,483 serving officers and staff.

In a police force where an officer's identity is a safety matter, the consequences were severe. One officer told the regulator they had spent more than £1,000 on CCTV and lighting at home. The Information Commissioner's Office fined the PSNI £750,000. Without the regulator's discount for public bodies it would have been £5.6m.

Lesson: a file is not what you see on screen. Hidden sheets, tracked changes, comments, attachments and earlier versions all travel with it.

2025: Meta's slides in the FTC antitrust trial

During the FTC's antitrust trial against Meta in April 2025, Meta filed exhibit slides carrying material that belonged to Apple, Google and Snap. The redactions were easy to remove: solid black boxes sat in the PDF as a separate layer over the content, and The Verge read what was underneath.

Lawyers for all three companies complained in open court about the botched redaction. Counsel for Apple and Snap called the errors egregious. Snap's lawyer described a "cavalier approach and casual disregard" toward the outside parties caught up in the case, and Apple's said it could no longer be confident Meta would protect its information. The Verge reported the exchange on 16 April 2025.

Lesson: the secrets you leak are often not yours. Third-party material in your own filing is your liability.

2025: the Justice Department's Epstein files

The Justice Department began releasing files on 19 December 2025 under the Epstein Files Transparency Act. Two separate redaction problems followed, and they are worth keeping apart.

The first was the familiar one. Readers found black boxes they could defeat with copy and paste. Snopes confirmed it in at least one document on 26 December 2025, and said it did not hold across the whole release. CNN traced that document to an old lawsuit filing that the department had republished with its faulty boxes intact.

The larger problem was omission. After a release of about three million pages in late January 2026, lawyers for survivors told ABC News on 30 January 2026 that victims' names had been published with no redaction at all, and that there were thousands of such errors. The department set up an email inbox for victims to report redaction problems. On 3 February, AFP reported that the survivors' lawyers had written to the judges asking for the files website to be taken down.

Lesson: at three million pages, redaction quality stops being a document problem and becomes a process problem.

Two failure modes behind almost all PDF redaction mistakes

1. Covered, not removed

A PDF stores text as instructions: draw these glyphs, at these coordinates, in this font. A drawn rectangle is another instruction. The viewer runs both, and the rectangle wins on screen because it is painted second. It loses everywhere else: copy and paste, text extraction, search indexing, screen readers, and any library that walks the content stream. Six of the nine cases here involve this mistake. In the Epstein release it was the smaller of two problems.

2. Metadata and hidden layers

The second mode is everything in the file that is not the page you are looking at. Hidden worksheets, as at the PSNI. Tracked changes, comments, attachments, thumbnails from an earlier draft. Document properties carrying the author, the organisation, the file path and the edit history. None of it shows when you scroll, and all of it ships when you send the file.

The Maxwell deposition sits outside both modes, and it is the useful edge case: structure that survives a correct redaction can still give the answer away.

How to check any redaction in 10 seconds

If it comes back, the redaction failed. This is the test reporters ran on the Manafort filing.

Passing it is necessary, not sufficient. It says nothing about metadata or hidden sheets, so also open the document properties panel and read the author, title, subject and keywords fields. And it will not catch a redaction that was never applied at all, which is what went wrong at Quinn Emanuel and across much of the Epstein release. For that you need a second reader.

The method every case here points at is the same: destroy the content instead of covering it, then verify the file you are about to send rather than the one you were looking at while you worked. Our walkthrough of how to redact a PDF without Adobe covers it step by step.

Why Redactron works the way it does

Redactron does not draw boxes. It rebuilds the exported PDF from flattened page images, so the redacted content is absent from the file rather than hidden behind a shape, and the original metadata is not carried over. In our own testing we byte-checked exports: 0 font objects, 0 text operators. After each export it re-opens the new file and checks it for text, fonts and the original metadata. That is a check of the output file, not proof that you marked everything, so review every page. It runs in your browser, so the document is never uploaded.

Run the copy-paste test on your own file

Redact a document, export it, then select all and paste. Nothing comes back, because there is no text layer left. Free for 3 documents a month; Pro is $39.99 once.

Redact a PDF free →

No signup. No upload. 3 documents free every month.