How to Redact a Court Filing Under Rule 5.2

Most privacy rules describe a standard and leave the application to judgement. Federal Rule of Civil Procedure 5.2 does the opposite. It names four categories of personal identifier and specifies, for each one, precisely how much may remain in the public filing. That precision is the whole reason redaction in litigation is harder than it looks, because three of the four categories are not deletions at all. They are partial disclosures, and a black bar over the entire value fails the rule just as surely as leaving it in.

The four categories, and what Rule 5.2(a) allows to survive in each. A social security number or taxpayer identification number keeps its last four digits. A birth date keeps the year, and nothing else. A minor's name is reduced to the minor's initials. A financial account number keeps its last four digits. Rule 49.1 of the Federal Rules of Criminal Procedure mirrors this on the criminal side, and most district courts add local requirements on top.

Read that list again with an eye on the shape of each answer, because they are four different operations. Two are partial masks that preserve a trailing fragment. One is a truncation that keeps a component of a date and discards the rest. One is an initialization that transforms a name into a shorter form derived from it. A tool that only knows how to remove text can do none of them correctly. It can only do the thing the rule does not ask for.

Over-redaction is a real failure and not a safe default. Rule 5.2(f) lets a party file the redacted version publicly and a complete unredacted copy under seal, which the court keeps as part of the record. Because that route exists, the public copy has no reason to hide more than the rule requires. Removing more than necessary invites a challenge from the other side, can delay a filing, and can oblige a re-filing on a clock you do not control. The instinct that more redaction is more safety is wrong here in both directions.

A few parts of the rule are worth knowing because they change whether any of this applies. Rule 5.2(b) exempts several categories outright, including the record of an administrative or agency proceeding, the official record of a state court proceeding, and a filing in a forfeiture proceeding. Rule 5.2(d) covers filings made under seal, which go in without redaction. Rule 5.2(e) lets a court, for good cause, order more redaction than the rule requires or limit remote electronic access by nonparties. And Rule 5.2(h) is the one people forget: a person waives the protection as to their own information by filing it unredacted and unsealed themselves.

Now the part that actually goes wrong. The most common failure is not a misreading of the rule, it is a redaction that was never real. Drawing a filled rectangle over text in a PDF viewer, a word processor or a browser editor leaves the original characters in the file. The shape sits on top of them. Select the page, copy, paste into a text editor, and the value comes back. Courts have unsealed this kind of mistake more than once, and it is a property of how the format works rather than a mistake by the person doing it. If your process ends with a black rectangle and a save, you have not redacted anything.

The second failure is volume, and it is the one that makes discovery productions dangerous. A pleading gets read carefully. The two hundred pages of exhibits attached to it do not. Inconsistency across a large production is worse than either treatment applied throughout, because a name redacted on page 40 and missed on page 900 demonstrates that review was attempted and failed, which is a harder position than never having claimed the review at all.

The third is the material that is not text. A scanned exhibit, a photographed document, a screenshot pasted into an appendix. There is no text layer to remove, so a text-based tool finds nothing and reports the page as clean. That is a silent miss, and it is the most expensive kind, because nothing about the output looks wrong.

Here is how Omit handles the four categories, since this is the workload the legal side of the product was built around. Policy is set per entity type, so a single configuration produces a filing that satisfies all four rules at once rather than requiring four passes. MASK keeps the trailing digits on a social security number, a taxpayer identification number and a financial account number. YEAR_ONLY reduces a date of birth to its four-digit year. INITIALS reduces a name to its initials for the minor's-name category. Those last two operators exist in the product for no other reason than that Rule 5.2 asks for those exact shapes, and no general-purpose redaction tool ships them.

On the failures above. PDF redaction removes the text objects rather than painting over them, and before the file is written the output is checked for whether the redacted value can still be recovered from what remains. If it can, the page is escalated to rasterize and flatten, which destroys the text layer outright. Scanned and photographed material goes through OCR, which reports where each word sits on the page, so the redaction box lands on the coordinates OCR gives rather than on a guess. Detection also covers legal-specific material beyond the four categories, including case numbers, Bates numbers and bar numbers, and privilege indicators can be flagged in the audit report so a reviewer sees them before a production goes out rather than after.

Two things worth stating plainly rather than letting you discover them. Omit reads PDF, Word, TXT, CSV, XLSX, JSON and image files. Word redaction reaches the headers, footers, comments, footnotes, tracked-change text and document properties as well as the body, which matters because those are where a matter name survives a pass that only ever looked at the page. RTF is not read yet. And an automated pass is a first pass. Rule 5.2 puts the responsibility on the filing party, not on a vendor, which is why the review step exists in the interface and why the audit report lists what was found and what was done to it. The tool's job is to make the review tractable across nine hundred pages, not to sign the filing.

Everything above runs on the machine doing the work. No upload, no API key, no account. For a filing that is under seal until the moment it is not, that is not a preference. It is the difference between a redaction workflow you can describe to a client and one you cannot.