Skip to content

Redact client and case data before it ever leaves the firm.

Omit keeps privileged material on the lawyer's machine while they draft, convert, and summarize.

Where sensitive data leaks out

Legal teams handle privileged client information, opposing-party data, and sealed material every day. Pasting a contract into a chatbot or emailing an unredacted exhibit can breach privilege and confidentiality obligations in a single click.

Rule 5.2 tells you exactly what to remove, and exactly what to leave

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, precisely how much may remain. That precision is why partial redaction matters more than removal in litigation. Blacking out a whole account number is not compliance with a rule that asks for the last four digits to be shown, and over-redaction invites its own motion. Rule 5.2 applies to every filing, including pleadings, motions, exhibits, transcripts and discovery material, and Rule 49.1 mirrors it on the criminal side. Responsibility sits with the filing party and nobody else, which is why an unredacted exhibit is an attorney problem rather than a vendor one.

Four categories, four different treatments

Social Security and taxpayer numbers keep their last four digits. Birth dates keep only the year. Minors are reduced to initials. Financial account numbers keep their last four digits. Four rules, and only one of them is plain deletion.

Omit ships an operator for each

MASK preserves the trailing digits on identifiers and account numbers. YEAR_ONLY reduces a date of birth to its year. INITIALS reduces a name to initials for minors. Because policy is set per entity type, one configuration produces a filing that matches the rule across all four categories at once.

Over-redaction is a failure too

Rule 5.2(f) lets a party file the unredacted version under seal, so the redacted copy has no reason to hide more than the rule requires. Removing more than necessary can draw a challenge, delay a filing, or oblige a re-filing on the clock.

India's DPDP Act exempts litigation from much of the law, but not from the security duty

The Digital Personal Data Protection Act, 2023 governs digital personal data processed in India, and the DPDP Rules notified in November 2025 bring most of its obligations into force by May 2027. Law firms get a partial carve-out: section 17(1)(a) disapplies much of the Act where processing is necessary to enforce a legal right or claim. It expressly keeps section 8(5), the duty to protect personal data with reasonable security safeguards, and that is the duty with the heaviest penalty in the Schedule, up to 250 crore rupees. Advisory and transactional work, from due diligence data rooms to employment investigations, may not sit inside the exemption at all. Either way, every client file, counterparty record and witness statement the firm holds needs protecting.

Security safeguards: nothing leaves the machine

Omit detects and redacts on the lawyer's own computer. Nothing is uploaded, so there is no vendor in the data flow, no cross-border transfer and no processor to bring under contract. The safeguard is how the software is built, not a promise in a policy.

Share only what the purpose needs

Minimisation in practice means redacting before a document goes to opposing counsel, a regulator, co-counsel or an AI tool. Omit detects Aadhaar numbers and Aadhaar virtual IDs, PAN, passport, voter ID, GSTIN, IFSC and UPI IDs alongside names, addresses, phone numbers and account numbers, and applies a set treatment to each. MASK can keep the last four digits of an Aadhaar number, the same form UIDAI uses for masked Aadhaar.

Pseudonymise when the work has to continue

When a matter needs the analysis but not the identities, consistent pseudonymisation gives each person the same placeholder across every document in a run. Vault-backed tokens can be reversed by authorised staff on the same machine, so identities are held back rather than lost.

A record of every run

Each run writes an audit report of what was found, where, and how it was treated. If the Data Protection Board asks what left the firm and in what form, the answer already exists, written at the time of the redaction.

The documents this actually applies to

Filings, exhibits and attachments

SSN, TAX_ID, DATE_OF_BIRTH, minors' names, ACCOUNT_NUMBER

The direct Rule 5.2 workload. Exhibits are the usual failure point, because the pleading gets reviewed carefully and the two hundred pages attached to it do not.

Discovery productions

PERSON, EMAIL, PHONE, ADDRESS, ORGANIZATION, ACCOUNT_NUMBER

Volume makes manual review impractical and inconsistency is the real risk: a name redacted on page 40 and missed on page 900 is worse than either treatment applied throughout, because it demonstrates that review was attempted and failed.

Privileged and work product material

PERSON, ORGANIZATION, matter and client references

Omit can flag privilege indicators in the audit report so a reviewer sees them before a production goes out. The flag is advisory: privilege is a legal determination and the tool does not make it for you.

Deposition and hearing transcripts

PERSON, ADDRESS, DATE_OF_BIRTH, ACCOUNT_NUMBER spoken aloud

Identifiers are read into the record constantly. Omit Voice transcribes locally and redacts the transcript, so a transcript that has to be filed can be prepared without sending it to a cloud service first.

One line, all four Rule 5.2 treatments

Before

Plaintiff Daniel Ashworth, SSN 412-88-7690, born 9 September 1981, maintains account 6642119083 at Northbank; his daughter Sophie Ashworth attends Redfield Primary.

After

Plaintiff Daniel Ashworth, SSN XXX-XX-7690, born 1981, maintains account XXXXXX9083 at Northbank; his daughter S.A. attends Redfield Primary.

Four categories, four different transformations, one policy. The plaintiff's own name stays because Rule 5.2 does not require adult party names to be removed. The last four digits survive on both numbers, the birth date becomes a year, and the minor becomes initials. A tool that blacks all of it out has not followed the rule, it has merely avoided the question.

A production log the court can read

Every run writes a report listing each detection, its type, its location and the operator applied. For a filing that is a record of exactly which identifiers were treated and how, produced at the time of the redaction. For a discovery production it is the consistency evidence: the same entity treated the same way across every document in the set. Privilege indicators are surfaced in the same report so they are seen before the production leaves, not after.

What records teams ask first

Does black box redaction actually remove the text?

In Omit, yes. It performs true PDF redaction, removing the underlying content from the file rather than drawing a rectangle over it. Drawing a shape is the cause of most published redaction failures, because the text underneath stays selectable and copyable in the filed document. If you have ever seen a redacted filing quoted verbatim in the press, this is usually why.

Can it keep the last four digits rather than removing the whole number?

Yes, that is what the MASK operator does, and it is the treatment Rule 5.2 actually requires for Social Security, taxpayer and financial account numbers. Policy is set per entity type, so account numbers can be masked while other identifiers are removed outright in the same pass.

How is privilege handled?

Omit flags likely privilege indicators in the audit report so a reviewer can look at them before a production goes out. It does not withhold documents on that basis and it does not make the call. Privilege is a legal determination, and a tool that claimed to make it would be overstating itself in a way that could cost you a waiver argument.

Does client data leave the firm?

No. Detection and redaction run on the machine that already holds the file. There is no upload, so there is no third party in possession of client confidences and no processor agreement to negotiate against your professional obligations.

Can we prove consistency across a large production?

The per-run report records every detection and its treatment, and consistent pseudonymisation maps a given entity to the same token throughout a run. Together those give you a defensible answer when opposing counsel asks why a name appears redacted in one document and not another.

What about state courts and other jurisdictions?

Rule 5.2 governs federal civil filings and Rule 49.1 the criminal equivalent. Many state courts adopt similar identifier rules with local variations, and some add categories such as home addresses. Because policy is configured per entity type, a firm can hold separate profiles per jurisdiction rather than relying on a reviewer to remember which rule applies.

Can lawyers use ChatGPT or Claude without pasting client data into it?

Yes. Protect, part of the Omit app and free for everyone, replaces names, numbers and other identifiers with labels as text is pasted, in any AI tool, so the model gets the substance of a clause without the client in it. It runs on the same machine and the same detection engine as file redaction.

How accurate is the detection?

Omit 1.0 runs a detection model we fine-tuned ourselves. On NVIDIA's public Nemotron-PII test set, which it never trained on, the model on its own misses 0.8% of personal data, down from 22.5% for the open model it was built from. On documents with no personal data it flags something in 12% of them, down from 76%, so reviewers dismiss far less. In the app, pattern and checksum recognizers run alongside the model, and every detection can be reviewed before the redaction is applied.

Does Omit help with India's DPDP Act?

With the parts a redaction tool can carry. Data never leaves the machine, which goes to the section 8(5) security duty that applies even to litigation work, and documents are redacted before they are shared, which is minimisation in practice. Omit detects Indian identifiers including Aadhaar, PAN, passport, voter ID and GSTIN. Consent, notices and data principal requests sit with the firm's own processes, and no tool makes a firm compliant on its own.

Further reading

  • Best Redaction Software, Compared Honestly

    Seven redaction tools compared on what they find by themselves, where your document goes to be checked, and what they cost to keep. Every competitor fact was read from the vendor's own site on 17 September 2026, and our own product's weak spots are listed too.

  • How to Redact a Court Filing Under Rule 5.2

    Federal Rule 5.2 names four categories of personal identifier and tells you exactly how much of each may remain. Only one of the four is plain deletion, which is why blacking everything out is not compliance. Here is the rule, the four treatments, and where filings actually fail.

  • ABA Opinion 512 and the Law Firm AI Policy

    Formal Opinion 512 does not ban generative AI for lawyers. It requires informed consent before client information goes into a self-learning tool, which turns a firm AI policy into a question about what leaves the machine. Here is what the opinion asks and how a policy answers it.

  • How to Redact a Word Document

    Highlighting text in black is not redaction, and neither is deleting it with track changes on. Omit removes the value from the file itself, across headers, footers, comments, footnotes, tracked deletions and document properties, entirely offline.

  • How to Unredact a PDF, and How to Prevent It

    Most recovered redactions come back the same trivial way, because a black rectangle is a drawing and the text is still underneath it. Here is why it happens, how to tell which kind you are holding, and what actually removes a value.

Local-only processing keeps privileged data inside the firm, which supports client-confidentiality duties and data-protection obligations without sending anything to a vendor.

Protect privilege by default.