Redacted Government Documents: What They Are and Why

Redaction is the process of erasing different kinds of words, images, or sections from this or that doc for good, before it is released to the public. It’s not when you cover the text with a black bar in a PDF viewer (which can sometimes still be copied or extracted). Solid redaction deletes the underlying data for good, so as a result, nobody can recover it. This distinction matters: several high-profile leaks in recent years, including cases tied to major federal investigations, have resulted from "cosmetic" redactions where blacked-out text remained selectable or extractable underneath the visual cover – most notably the December 2025 Epstein files release, where some redactions could be defeated with a simple copy-paste.

Redaction is applied to public records requested through formal channels, most commonly the Freedom of Information Act (FOIA), as well as state-level public records laws, court filings, and congressional document productions.

The law that makes all this possible is the Freedom of Information Act – FOIA for short, officially found at 5 U.S.C. § 552\. It gives regular people the legal right to ask federal agencies for their records. Once you file a request, the agency's supposed to respond within 20 business days, though in practice they can (and often do) claim "unusual circumstances" to buy themselves more time.

From there, the agency goes hunting for anything relevant to your request, then reviews it all to figure out what's fair game to hand over – and what has to be held back or blacked out under one of FOIA's nine built-in exemptions.

Many states have parallel laws (often called "sunshine laws") that apply similar logic to state and local government records, though the specific exemptions vary by jurisdiction.

When a FOIA release comes back, each black bar usually carries a code identifying which exemption the agency invoked – "(b)(1)" for classified national security material, "(b)(6)" for personal privacy, "(b)(7)" for law enforcement records, and so on, following the exemption's position in 5 U.S.C. § 552(b). Two things to know when reading them: the same information should carry the same code throughout a response – a name redacted under (b)(6) on page one carries (b)(6) on page one hundred – and declassified historical records use a different marking system entirely, based on Executive Order classification categories rather than FOIA exemptions. Even after the 25-, 50-, and 75-year declassification milestones, material that would still harm personal privacy or national security stays blacked out.

Here’s a typical page from a FOIA production – each black bar is labeled with the exemption the agency invoked to withhold it:

There are several distinct reasons an agency redacts a document rather than releasing it in full:

The most sensitive category covers classified material: military operations, intelligence sources and methods, diplomatic communications, and weapons programs. Disclosure here can endanger both people and ongoing operations, which is why Exemption 1 redactions are typically the least negotiable.

A lot of documents end up with random personal stuff mixed in (Social Security numbers, medical details, home addresses, that kind of thing) belonging to everyday people who just happen to show up in the paperwork. Blacking that out protects those folks, since they're not who the request is actually about.

Law enforcement and regulatory agencies often withhold details that could tip off suspects, compromise informants, or interfere with active cases.

When companies submit information to regulators, that data can include confidential formulas, financial figures, or negotiating positions that would cause competitive harm if disclosed.

Internal agency drafts, recommendations, and pre-decisional discussions are often shielded so that officials can debate policy candidly before a final decision is made public.

Together, these categories are often referred to under the general umbrella of sensitive government information – a term covering anything from classified intelligence to a private citizen's medical file swept up in a larger investigation.

The volume of confidential government documents processed each year has grown dramatically, and so has the share of material withheld or redacted.

• In Fiscal Year 2025, federal agencies received a record-setting 1,707,197 FOIA requests – a 13.7% increase over the previous fiscal year and the highest number ever recorded. Agencies processed 1,635,055 of them.
• This marks the third consecutive year that FOIA requests have reached seven figures, and the number of requests received has increased every year since fiscal year 2020.
• Despite dealing with processing more requests than ever before, full disclosure is still just an exception, not the rule. During fiscal year 2024, just 12.1% of processed FOIA requests were granted 100% fully. It means that most of the releases involved partial redaction or outright denial.
• Backlogs have grown alongside demand. Fiscal year 2024 marked the third consecutive year that backlogged cases topped 200,000, and administrative appeals jumped 59.38% year-over-year as requesters pushed back on denials and heavy redactions.
• Staffing has not kept pace with demand. According to the Justice Department's own FY2025 summary, just 4,823 full-time employees administer FOIA across the entire federal government, at an estimated cost of $661 million – meaning each employee is responsible for well over 300 requests a year.

These numbers illustrate a central tension in transparency policy: as public interest in public records grows, so does the administrative strain on the agencies responsible for reviewing and redacting them, which in turn increases both backlogs and the likelihood that release decisions default toward caution and heavier redaction.

Given the sheer volume of pages agencies must review line-by-line for exempted material, many agencies have begun adopting AI-assisted redaction tools. The ATF, for example, has credited automated redaction software with meaningful productivity gains in FOIA processing, and agencies like ARPA-H have explicitly sought AI tools to handle redaction and document review at scale. The pitch from people who back automated redaction is pretty simple: let the software handle the boring, repetitive stuff (like spotting Social Security numbers or other personal identifiers) so it can turn a request that used to take weeks into one that takes hours – and free human reviewers for the judgment calls that genuinely require a human brain, like deciding whether a passage is sensitive enough to count as a national-security exemption.

At the same time, critics tend to say that automation comes with a bunch of risks – the question of whether AI redaction is safe enough for sensitive documents is far from settled. The latter usually include over-redaction of releasable material, under-redaction of sensitive data, and the same "cosmetic redaction" failures that have caused real-world leaks when automated tools mark text visually but fail to handle the underlying file. And that’s huge.

That gap is why agencies handling routine FOIA production increasingly move to dedicated government redaction software rather than manual black-boxing: the tool decides whether the underlying data is actually destroyed, not the reviewer drawing the box.

Redaction exists at the intersection of two legitimate public interests: the public's right to oversee its government, and the government's obligation to protect national security, individual privacy, and the integrity of ongoing legal processes. Neither interest is absolute, which is why FOIA and similar laws build in structured exemptions rather than blanket secrecy or blanket disclosure.

For requesters, understanding these categories can make the redaction process less opaque. A document that comes back with entire pages blacked out is not necessarily evidence of concealment for its own sake – it may reflect a genuine legal exemption. But the process is far from an exact science. The National Security Archive has documented hundreds of "dubious secrets" – cases where censors blacked out documents that had already been released in full, or redacted entirely different parts of the same document at different times. In one example, four reviewers examined the same State Department memo on Rwanda over a 12-year period and reached four different conclusions: one withheld the final paragraph as classified (B1), a later reviewer released it but redacted a different section as privileged deliberation (B5), and a third release restored that section while deleting the final paragraph all over again. At the same time, the growing backlog and declining full-grant rate reflect real strain on the system, and have fueled ongoing debate among transparency advocates, journalists, and lawmakers about whether current practices strike the right balance.

To cut a long story short, redacted government documents are a routine and legally structured part of how the U.S. government balances transparency with security and privacy. Whether the underlying material involves classified documents tied to national defense or simply the private details of a citizen swept into an unrelated file, redaction serves a defined legal purpose under FOIA's nine exemptions. But as request volumes climb past 1.7 million a year and full-grant rates hover near 12%, the practical reality of redaction (its scale, its backlogs, and its increasing reliance on automated tools) is becoming as important a story as the underlying documents themselves.