Famous Redaction Failures: When the Black Box Lied

By the SladdPDF team · Published August 23, 2026 · Updated August 23, 2026

Redaction failures happen when a document looks censored on screen but still carries the secret text inside the file. In the seven cases below — from the US Justice Department in 2003 to a Swedish inspectorate in 2025 — the hidden words were recovered by selecting the black bars and pressing Ctrl+C, by switching Word's tracked changes back on, or by reading an index nobody thought to redact. The pattern never changes: covering text is not the same as removing it.

Why do redaction failures keep happening?

Redaction failures keep happening because every common tool makes it easy to cover text and hard to remove it. A PDF page is not a picture. It is a list of objects: text objects with their characters and coordinates, images, vector shapes. When you drag a black rectangle over a name, you add one more shape to the list. Nothing is deleted. The name still sits underneath at the same coordinates, and every select tool, search index and text-extraction library can still read it.

The result is a document that lies. It looks censored to a human eye and stays fully legible to software. Two keystrokes — Ctrl+A, Ctrl+C — undo the whole job, which is exactly what journalists have done to government agencies, defence ministries and elite law firms for more than twenty years.

Real redaction removes the underlying content. The strongest form goes further and rasterises the page — flattening text, shapes and layers into a single flat image, so there is no text object left to recover. That is the difference between the cases below and a document that holds up.

2003: the Justice Department's black highlighter

The first famous PDF redaction failure was published by the institution that should have known best. In October 2003, the US Department of Justice released a 186-page internal study of diversity among its own attorneys under the Freedom of Information Act — with roughly half the text blacked out, citing the FOIA exemption that protects internal deliberations. Whole sections on stereotyping, racial and gender tension, harassment and mentoring were covered.

Technical cause: the document had been sanitised in a word processor by setting the highlighter colour to black, then published as a PDF. The text layer was untouched, so the black bands could simply be selected and copied out with the free Acrobat Reader.

Consequence: a transparency website called The Memory Hole republished the report with the black rectangles stripped away, and the Justice Department found itself publicly defending its editing choices instead of discussing the study. Two decades later, it is still the canonical FOIA redaction cautionary tale.

2005: the UN report that kept its own draft

Not every leak comes from a black box — some come from the file's memory. In October 2005 the UN commission investigating the assassination of Lebanese prime minister Rafik Hariri, led by German prosecutor Detlev Mehlis, released its report as a Microsoft Word document with tracked changes intact.

Technical cause: metadata, not overlay. Reporters who reversed the tracked changes could read an earlier draft in which named senior officials — including Maher al-Assad, the Syrian president's brother, and military-intelligence chief Assef Shawkat, alongside Hassan Khalil, Bahjat Suleyman and Jamil al-Sayyed — appeared in a passage that the published version had softened to a generic reference to senior officials.

Consequence: a diplomatic row in which the deletions became the story. Mehlis said he had removed the names himself, on a presumption of innocence and under no pressure from any party. The case is the standard reminder that a file can leak the words you deleted as easily as the words you covered.

2009: the TSA screening manual

In December 2009 the US Transportation Security Administration posted a copy of its airport screening operating manual to the Federal Business Opportunities website as part of a contract solicitation. A contract employee had redacted the 93-page document by drawing black boxes over the sensitive passages.

Technical cause: overlay. The text under the boxes was never removed, so standard cut-and-paste exposed screening procedures, the special handling rules for CIA officers, diplomats and law-enforcement personnel, and the technical settings and tolerances of the metal and explosive detectors used at checkpoints.

Consequence: the TSA said the document had been improperly posted, pulled it, opened an Office of Inspection review and placed the individuals involved on administrative leave. None of that mattered: the file had already been copied to mirrors, where it remains. It is the case security professionals still reach for first when explaining why the copy-paste test exists.

2011: British nuclear submarine secrets

In April 2011 the UK Ministry of Defence published safety regulators' advice on the propulsion plant for the Successor SSBN — Britain's next generation of ballistic-missile submarines — on the Parliament website, following Freedom of Information requests.

Technical cause: overlay again. The department had turned the background of the sensitive passages black rather than deleting them, so copying the blacked-out sections into another document revealed assessments of vulnerabilities in British submarines, including expert commentary on how much structural failure could lead to a reactor core meltdown on board.

Consequence: the MoD replaced the file with a properly protected version, but search engines had already indexed the contents as plain searchable text. The episode drew international coverage as a textbook blunder — and it was not an isolated one, with further poorly protected documents surfacing on UK government sites later that year.

2019: the Manafort filing that lasted minutes

On 8 January 2019, lawyers for Paul Manafort, former chairman of the Trump presidential campaign, filed a response to the Special Counsel's office in federal court with black bars over the disputed passages. Reporters had the hidden text within minutes.

Technical cause: overlay. The redaction bars were drawn over the text but never flattened or sanitised, so the covered passages could be copied straight out of the filing and pasted into a fresh document.

Consequence: the filing revealed that Manafort had shared 2016 campaign polling data with Konstantin Kilimnik, an associate the Special Counsel's office said had ties to a Russian intelligence service, along with a meeting in Madrid and discussion of a Ukraine peace plan. It became one of the most consequential disclosures of the Mueller investigation — and, as the American Bar Association's own write-up of embarrassing redaction failures notes, a permanent fixture of legal-technology training.

2020: a deposition cracked by its own index

The Ghislaine Maxwell case shows that redaction can fail even when the black bars hold. In October 2020 a federal court released Maxwell's 2016 deposition from Virginia Giuffre's defamation suit, with the names of prominent men blacked out — among them Bill Clinton, Alan Dershowitz and Prince Andrew.

Technical cause: collateral data. The transcript shipped with the court reporter's complete alphabetised index of every word in the document, including the redacted ones, with page and line references attached. Journalists at Slate reconstructed the hidden names from where each redacted word fell in the alphabet and how the index lined up with the bars.

Consequence: the reconstruction was published within a day of the release. The lesson generalises far beyond one deposition: redaction has to cover every derived copy of the text — indexes, tables of contents, bookmarks, OCR layers and appendices — not just the visible body.

2024–25: the masking tool that removed the masking

The most uncomfortable case is the most recent, because the organisation was using dedicated redaction software. Sweden's health and social care inspectorate, Inspektionen för vård och omsorg (IVO), released documents in December 2024 concerning families subject to social-services interventions. Names and addresses that should have been protected by confidentiality were in clear text.

Technical cause: process and tooling. IVO's own digital masking tool removed the masking during document handling — which the agency described as an unfortunate combination of a known defect in the technical system and human error.

Consequence: IVO reported itself to the Swedish data protection authority, Integritetsskyddsmyndigheten. Public-service broadcaster Sveriges Radio's Ekot programme, which received the documents, showed it was not a one-off, and reported that internal documents had flagged a significant risk of mishandling while routines were tightened only after journalists started asking questions.

What do these redaction failures have in common?

Every case above comes down to one confusion — hiding text instead of deleting it — and to a missing verification step at the end. Here is the pattern in one view.

Four lessons fall out of the table:

• Redaction failures are a workflow problem, not a competence problem. The list includes a justice ministry, a UN investigation, an aviation-security agency, a defence ministry, a top-tier defence team and a national inspectorate. Careful people using the wrong step produce the same result every time.
• The text you never looked at is as dangerous as the text you covered. Revision history, document metadata, OCR layers and indexes all carry the same words in another place.
• Trusting the tool is not the same as verifying the output. IVO's software was built for masking and still shipped clear text. The only meaningful check is on the exported file, not on the screen preview.
• Once it is out, it is out. The TSA manual and the MoD report were both withdrawn within days, and both are still circulating. Redaction has exactly one chance to work.

A pre-publication redaction checklist

Run this list on the exported file — the copy you are about to send — every single time.

1. Delete, do not cover. Use a tool that removes the content rather than drawing over it. If you cannot tell which one you have, assume the worst.
2. Flatten the page. Rasterising each page turns text, layers and hidden objects into pixels, which removes the recovery path the 2003, 2009, 2011 and 2019 cases all relied on.
3. Do the copy-paste test. Open the export, press Ctrl+A, Ctrl+C, and paste into a plain text editor. Anything sensitive that appears means the job failed.
4. Search for what you removed. Use the viewer's find function on two or three of the redacted names or numbers.
5. Check the derived text. Indexes, appendices, tables of contents, bookmarks and thumbnails can repeat what the body hides — the Maxwell mistake.
6. Strip metadata. Title, author, creation tool and hidden tags travel with the file. Word and PDF revision data is a documented leak route, not a theoretical one.
7. Check for collateral clues. Bar length, page numbering, file names and surrounding sentences can identify what is behind the bar.
8. Keep the master separate. Redact a copy, archive the original somewhere the release process cannot reach.

If you want to run steps 1, 2 and 6 in one pass, you can redact a PDF in your browser with SladdPDF: draw over what should disappear, choose secure mode so each page is rasterised on export, and tick the metadata-removal option. The file is processed locally with JavaScript and WebAssembly and is never uploaded to a server, which also means there is no copy of your sensitive document sitting in someone else's storage. For the underlying mechanics, see our explainer on why a black box is not redaction.