Anonymize a judgment or court documents: a guide for lawyers and legal professionals

Why anonymize a judgment or a court document

A law firm or legal department constantly handles court decisions and case documents: judgments, written submissions, official reports, expert reports. These documents almost always contain personal data about the parties, sometimes about third parties, and often information covered by the lawyer's professional secrecy. Anonymizing them becomes necessary as soon as they leave the strict framework of the proceedings: publication in a journal, use in internal training, sharing with a colleague on a similar case, or simply analysis by an external tool, including an AI. The official reports and writs of commissaires de justice raise the same question.

This anonymization must be rigorous: a poorly anonymized court decision can expose a victim, a minor, or a person in a sensitive medical or financial situation, with very real consequences for them.

The French framework: open data for court decisions

In France, the 2016 Law for a Digital Republic established the principle of making court decisions available to the public. Law No. 2019-222 of March 23, 2019 on programming for 2018-2022 and reform for the justice system set the redaction rules, laid down in Article L111-13 of the Code of Judicial Organization for the ordinary courts and in Article L10 of the Code of Administrative Justice for the administrative courts:

• the surnames and first names of individuals mentioned in the decision, when they are parties or third parties, are systematically redacted before the decision is made available to the public;
• any other element that makes it possible to identify the parties, third parties, and also judges and court registry staff, is also redacted when its disclosure is likely to undermine the safety or privacy of these people or those close to them;
• the identity data of judges and court registry staff may not be reused to evaluate, analyze, compare or predict their actual or alleged professional practices: such reuse is a criminal offense.

This framework concerns first and foremost official publication organized by the State (public case-law databases). It does not relieve a law firm of its own vigilance whenever it distributes, publishes or shares a decision or a document outside the strict procedural framework: the GDPR then applies in full, with its obligations of data minimization and, where applicable, security of processing.

Redaction of court decisions: the two levels, the timetable, the engine

Decree No. 2020-797 of June 29, 2020, issued to implement Article 33 of the Law of March 23, 2019, organizes how decisions are made available. The Cour de cassation (France's supreme court for civil and criminal matters) is responsible for this for the ordinary courts, and the Conseil d'État for the administrative courts. Redaction happens at two levels:

• Systematic redaction: the surnames and first names of individuals who are parties or third parties, in all published decisions.
• Additional redaction: any other element that would make it possible to identify the parties, third parties, judges or court registry staff (address, date of birth, rare occupation, precise location), when its disclosure would undermine their safety or privacy. It is decided case by case.

The open data timetable

An order of April 28, 2021 sets the dates of availability, court by court. It has been amended several times, most recently by an order of August 29, 2025 that postponed the labor courts and criminal matters:

Open data only covers decisions handed down after these dates: older decisions are not put online at the same pace.

The Cour de cassation's redaction engine

To handle these volumes, the Cour de cassation developed, with the AI Lab of the interministerial digital directorate (DINUM), machine-learning pseudonymization software, in production since late 2019 according to Acteurs publics. It first spots surnames and first names, then other identifiers such as license plates, e-mail addresses or social security numbers. Its rules keep the logical links between people, which simple redaction does not allow. Staff review and annotate the decisions, and these annotations are used to improve the model.

A firm can apply the same logic to its own documents: one stable label per person, [PERSONNE_1] for the claimant, [PERSONNE_2] for the defendant, keeps the reasoning readable without exposing anyone. This is what AnonymFiles does on the computer, across an entire file.

What to redact in a decision or a document

In French open data, systematic redaction covers the surnames and first names of individuals who are parties or third parties, such as witnesses. Other identifying elements are redacted only if their disclosure threatens the safety or privacy of the people concerned or those close to them, and this rule also applies to judges and court registry staff: their names are not redacted as a matter of principle, but reusing them to evaluate or predict their practices is prohibited.

Beyond names, an attentive legal professional should also look for:

• Postal addresses and precise locations that, combined with other elements, make it possible to trace a person.
• Identification numbers: case number, social security number, license plate, IBAN mentioned in the documents.
• Very specific medical or financial details that, even without an associated name, may be enough to re-identify a person within a small circle (a small village, a company with a few employees).
• Contact details: e-mails and phone numbers appearing in the exchanges filed with the case.

The test to apply is simple: could someone who knows a little about the context of the case guess who the parties are by reading the redacted document? If so, the anonymization is not sufficient.

Step-by-step method for anonymizing a document

1. Work on a copy, never on the original filed with the case or archived by the firm.
2. List the parties and third parties named in the document: claimant, defendant, witnesses, experts, with all their spelling variants (surname alone, surname and first name, title).
3. Replace each occurrence with a stable label: the same person must carry the same label everywhere in the document, for example “[PERSONNE_1]” for the claimant, “[PERSONNE_2]” for the defendant, which keeps the legal reasoning readable without exposing the real identity.
4. Truly redact sensitive passages, do not just cover them with a visual rectangle. Adobe Acrobat Pro has a dedicated redaction tool that deletes the underlying text; Preview on Mac (Tools › Redact) and LibreOffice Draw (Tools › Redact) also offer a real, free redaction tool. Then check the final file: Cmd+A or Ctrl+A, copy and paste into a text editor.
5. Check the file's metadata (author, company, revision history), which can reveal names even when the body text is clean. See how to clean up metadata.
6. Reread the anonymized document in full, including appendices and attachments, before any distribution or sending to a third party.

For a single file, this manual method with Adobe Acrobat Pro, Preview or LibreOffice works, provided you take the time needed for a full review. It quickly becomes cumbersome when several documents from the same file, or several similar files, have to be handled.

Anonymizing an entire file with AnonymFiles

AnonymFiles is a desktop app designed for this kind of volume, coming soon for Apple silicon Macs and Windows. Once installed, it works offline: no account, no online license check, no telemetry, no cloud chat — a particularly relevant point for documents covered by the lawyer's professional secrecy, which should never pass through a service whose hosting you do not control. See also AnonymFiles for law firms.

You drop in the case file, subfolders included (PDFs, even scanned, Word .docx, e-mails, spreadsheets, presentations, images, .txt and .md): the app detects people, organizations, addresses, e-mails, phone numbers and IBANs, then replaces them with stable labels such as [PERSONNE_1], [ORGANISATION_1], the same across every document. You then compare each original with its copy side by side, redact by hand whatever else needs it and approve document by document: nothing is exported without your approval. The export produces copies with neutral file names, in the same folder structure, with an index of exported and excluded documents; the mapping table between labels and real identities stays on your computer and is never attached to the copies. The originals are never modified.

Two things to know: scanned documents (digitized official reports, ID documents, signed letters) are read through local OCR, and a stamp or handwritten note that the OCR cannot read is masked in one click during review (see our methods for scanned documents). And detection does not replace the lawyer's review. On the AI side, the local MCP server exposes only the redacted, approved copies to Claude Code, Codex or Cowork.

AI, professional secrecy and court documents

More and more legal professionals use a generative AI to summarize a decision, prepare an oral argument or analyze a large volume of documents. The right reflex is the same as for any other confidential document: anonymize before sending, never after, and never by handing the anonymization itself to the AI, which would mean sending it the sensitive data before checking that it may receive them.

Here the lawyer's professional secrecy adds to the GDPR question: a document covered by this secrecy should not pass through a third-party service without strict control over what is actually sent. Processing the document locally, with a tool that stays on the workstation, before any use of an AI assistant, keeps you in control of what leaves the firm, provided you choose local processing that you can verify. The same requirement applies to reports and petitions in insolvency proceedings.

Pitfalls to avoid

• Do not consider redacting the parties' names sufficient if the context of the case (place, date, occupation, unique details) still makes it possible to identify the person.
• Do not reuse a document already “anonymized” by a third party without checking the quality of the work yourself: insufficient redaction engages your liability when you republish or send it.
• Do not forget appendices, attachments and earlier versions of a document, which often contain the most sensitive information.
• Never consider any automatic tool, whatever it is, to be 100% reliable: a human review remains essential before any distribution.

To clarify the vocabulary used in this article, the distinction between anonymization and pseudonymization under the GDPR is a useful starting point. If the document is intended for an AI, our guide to anonymizing a document before sending it to ChatGPT or an AI details the method. And for a reliable method of redacting a PDF in general, see redacting a PDF step by step.