Lawyers: anonymizing case files and exhibits without losing meaning

To anonymize a case file without losing meaning, replace each identifier with a consistent label. Jane Example becomes Party A or [NAME1] everywhere: in the pleadings, the exhibits and the letters. Then check what the context still gives away. Masking names reduces exposure. It does not lift your duty of confidentiality or protect privilege by itself.

Lawyers share documents daily: a draft for an expert, an exhibit for the other side, an old brief for training, a contract for an AI tool. Each time, ask what the reader can still learn about the client. Texts are quoted as of September 2026. This is not legal advice.

What do the bar rules say about confidentiality?

The rules protect information, not just names. First, the client's name is itself protected: the French national bar rules (RIN) list it, and the German Federal Bar (BRAK) wrote in December 2024 that even the existence of a mandate is covered. Second, removing names may not be enough. The comment to ABA Rule 1.6 (US) allows a hypothetical only if the listener is not reasonably likely to work out the client or the situation.

What should you mask in pleadings, exhibits and letters?

Start with direct identifiers. They hide where people forget to look: the caption, the signature block, the footer, the list of exhibits.

• Names of parties, witnesses, experts, judges and opposing counsel, with initials.
• Addresses, emails, phone numbers, dates and places of birth.
• Docket numbers, your firm's file reference, policy and contract numbers.
• Official identifiers: ID, passport, social security, tax, company and VAT numbers.
• Bank details: IBAN, BIC, card numbers, and unusual amounts.
• Land registry references, vehicle plates, property addresses.
• Letterheads, logos, stamps, signatures, handwritten notes.

Then look for indirect clues. In its AI guide of September 2024, the French National Bar Council (CNB) warns that amounts, internal project names, or a drafting style can identify a client once combined. In US federal civil filings, unless the court orders otherwise, Rule 5.2(a) of the Federal Rules of Civil Procedure allows only the last four digits of social security, taxpayer and financial account numbers, the year of birth, and a minor's initials.

How do you keep the text readable after masking?

Black bars everywhere destroy a legal text. The reader can no longer tell who sued whom. Use labels instead: one person, one label, across the file. The CNB gives the same advice: replace client names with fictitious initials such as X or Y, and a street address with a wider area such as a city.

When roles matter, use the role as the label: Claimant, Employer, Expert 1. Keep the key, the list linking each label to the real value, in a separate file inside the firm. Rule 5.2(g) formalizes this for US federal civil filings: a redacted filing may come with a sealed reference list giving each item a unique identifier.

How do you handle exhibits: scans, stamps, signatures, handwriting?

Exhibits are the hard part. Many are scans, so the page is a picture. A tool must first read it with OCR, software that recognizes printed characters. OCRmyPDF, an open-source OCR tool, says in its documentation that its engine cannot recognize handwriting and that poor scans give poor results. Never rely on automatic detection alone.

• Stamps and seals: they often carry a name, a date and a number. Mask the whole stamp.
• Signatures and initials: they identify a person as surely as a name.
• Handwritten notes in the margin: read them yourself. Software usually will not.
• Photos, ID copies and screenshots: check faces, plates and numbers.

Make sure nothing survives under the masks. The UK Information Commissioner's Office (ICO) says text only covered by a shape stays in the electronic file. Adobe's documentation says the Redact tool in Acrobat Pro permanently removes content, and that sanitizing removes comments, metadata or hidden layers. Another route is a flattened export, where each page becomes an image. ONYRI Sanitize works this way, in the browser, without uploading the file. It does not read handwriting either, though you can mask it by hand with the “Draw an area” tool.

Can you give a masked file to an AI tool?

Masking helps, but bar guides stay cautious. The CCBE guide on generative AI (October 2025) says lawyers are responsible for what they enter as a prompt, and that the provider may store and reuse it unless stipulated otherwise. BRAK (Germany) advises abstract prompts and, where a document must be uploaded, full anonymization first if possible. The CNB (France) presents pseudonymisation as good practice, but not for every task.

How do you verify the result and keep a record?

1. Work on a copy, never the original, as the ICO advises.
2. Search the final file for each masked value: name, initials, last digits of numbers.
3. Select all, copy, paste into a plain text editor. No masked value should appear.
4. Check file properties, comments, tracked changes and the file name.
5. Read the exhibits page by page: stamps, signatures, handwriting.
6. Ask a colleague who does not know the case: can they guess who it is?

Keep a record. The ICO asks organizations to record who redacted the information, when and why, and to keep clearly labeled redacted and unredacted versions. Store the original, the shared copy and the label key in the client file.

Frequently asked questions

Sources & references