90,232 Files Removed: What the DOJ Took Down and Why It Matters

The Department of Justice has removed 90,232 files from its public Epstein database since the initial release on January 30, 2026. CBS News first reported in late February that more than 47,000 files had been taken down. Our independent monitoring puts the actual number nearly twice as high.

We know the exact number because we have been tracking it. Our DOJ Audit system monitors the justice.gov/epstein database continuously, logging every removal, modification, and size change. We maintain SHA-256 cryptographic hashes for 1.38 million files, allowing us to verify not just what was removed, but whether any files were altered before being taken down.

Of the 90,232 removed files, we have preserved copies or verification hashes for 90,230 of them. That is a 99.997% preservation rate. Two files, out of more than 90,000, are unaccounted for in our archive.

This article presents what the data shows.

The Numbers

The removals span 11 of the 12 data sets released under the Epstein Files Transparency Act. Data Set 7 is the only set with no documented removals.

Data Set 9 accounts for 97.8% of all removals. This is not a coincidence. DS9 contains the email evidence, internal DOJ correspondence about the 2008 non-prosecution agreement, and communications between Epstein's associates and legal team. It is the single most sensitive data set in the collection.

What Was Removed

The removed files fall into several categories, and not all removals are equal.

31,676 image-only files. The largest single category. Some of these contained unredacted photographs of victims, which the DOJ has acknowledged should never have been published. Attorneys representing nearly 100 survivors filed declarations stating that the exposure of these images was "life threatening." Eight women reported receiving death threats after their photographs were published without redaction. The removal of these files is defensible and necessary.

25,054 files categorized as "other." This is the second-largest category and the most opaque. The "other" designation provides no information about what these documents contain or why they were removed. Some may contain victim information. Some may not. Without access to the underlying documents, independent verification is impossible.

7,578 correspondence files. Emails, letters, and memos. Some of these files contained personal information about victims or witnesses. Others contained communications between Epstein, his associates, and individuals in positions of political or financial power. The DOJ has not distinguished between removals motivated by privacy and removals motivated by other concerns.

5,679 financial records. Bank statements, wire transfers, invoices, and accounting documents. The financial records are critical to understanding Epstein's network of influence and the movement of money between his entities. Our Forensic Finance section traces $5.3 billion through 224 entities and 446 financial flows, much of it sourced from documents in this category.

1,365 law enforcement records. Investigative reports, case notes, and FBI documentation. These are primary source materials from the federal investigation. Their removal from public access directly impairs the ability of journalists, researchers, and victims' attorneys to evaluate the adequacy of the investigation.

1,357 prosecution records. Materials related to charging decisions, plea negotiations, and the 2008 non-prosecution agreement. These documents are central to the public interest question of why Epstein received such favorable treatment from federal prosecutors.

The Privacy Defense

Deputy Attorney General Todd Blanche stated that the DOJ removed documents containing "explicit images or survivor information, including one document with unredacted photos of 21 survivors along with most of their birthdates." Lawyers for the DOJ told a federal court that files would be re-posted "ideally within 24 to 36 hours" after evaluation and proper redaction.

That was weeks ago. The vast majority of the 90,232 files have not been re-posted.

The privacy defense is valid for some removals. We have conducted our own redaction of victim PII across our database, removing 16,924 instances of personally identifiable information from 1,567 documents. We take victim privacy seriously.

But the privacy defense does not account for the removal of fully redacted call logs, photographs of jail furnishings, administrative documents with no victim connection, and the tens of thousands of files in the "other" and "uncategorized" designations that the DOJ has not explained.

What We Preserved

Our Document Integrity system provides three layers of verification:

SHA-256 hashes. We maintain cryptographic hashes for 1.38 million files sourced from independent community archives. These hashes allow us to verify that any document returned to the database matches what was originally published.

OCR text. For 1,960 of the removed files, we have preserved the full text content through optical character recognition. This includes 819 correspondence files, 339 financial records, 258 phone records, and 53 law enforcement documents.

Person links. 1,276 of the removed files are linked to specific individuals in our person database. This means we can identify which persons are most affected by the removals and assess whether the pattern of removals correlates with the prominence or political sensitivity of the individuals involved.

The Pattern

The data does not support the conclusion that all 90,232 removals were motivated by victim privacy.

The category breakdown shows that image files, the category most likely to contain victim photographs, account for only 35% of removals. Correspondence, financial records, law enforcement documents, prosecution materials, and uncategorized files together make up the remaining 65%.

The concentration in Data Set 9, which contains the most politically sensitive materials in the collection, is suggestive. DS9 houses the email evidence that connects Epstein to current and former government officials, financial executives, and public figures. It contains the documentation of the 2008 non-prosecution agreement, which remains one of the most controversial decisions in the history of federal criminal law.

We do not claim that the removals were politically motivated. We present the data and let it speak.

The Timeline

The removals did not happen all at once. They occurred in waves, each triggered by a different public event.

The first wave came within days of the January 30 release, after victims' attorneys filed emergency motions documenting the exposure of nude photographs with unredacted faces. The DOJ pulled thousands of image files across multiple data sets. This response, while belated, addressed a genuine and urgent privacy failure.

The second wave followed media reports identifying specific named individuals in the released documents. Thousands of additional files were removed across correspondence, financial, and law enforcement categories. The DOJ did not issue public explanations for these removals.

The third wave coincided with congressional pressure. As the House Oversight Committee escalated its investigation in February and March, additional files were taken down. The total count crossed 90,000 during this period.

Each wave reduced the publicly available portion of the Epstein archive. The DOJ originally claimed to have released more than 3 million pages. The current publicly accessible total is approximately 2.7 million pages, a reduction of roughly 10%.

Independent Verification

The value of independent document preservation cannot be overstated. The DOJ controls the primary copy of the Epstein archive. Without independent copies and cryptographic verification, there would be no way to assess the scope of removals, detect alterations, or verify that restored documents match their original versions.

Our preservation system operates on three principles. First, we store SHA-256 hashes sourced from independent community archives, not from the DOJ itself. This means our verification data cannot be retroactively altered by the same institution that controls the source documents. Second, we extract and index OCR text independently, preserving the informational content of documents even when the source files become unavailable. Third, we link documents to persons, creating a relational map of the archive that survives individual file removals.

Other community projects have contributed to this effort. The rhowardstone/Epstein-research repository maintains a complete EFTA page inventory. The beak2825 archive tracks server responses and file modifications. Multiple torrent archives ensure distributed preservation of the raw files.

What Needs to Happen

The House Oversight Committee has subpoenaed Attorney General Bondi to testify about these issues. That is a necessary step but not a sufficient one.

What is needed is a public accounting, file by file, of why each of the 90,232 documents was removed, what category of exemption applies, and when each file will be restored to the public database. The Epstein Files Transparency Act requires disclosure. Removal without explanation is not consistent with that mandate.

Our DOJ Audit page is updated continuously. Anyone can view the full dataset of removed files, filter by data set and category, and assess the scope of the removals independently.

The files belong to the public. The question is whether the Department of Justice agrees.

Key Resources

This article is based on data collected through automated monitoring of the DOJ Epstein Library at justice.gov/epstein. All statistics cited are derived from our database and can be independently verified through the resources linked above.