Anonymize legal documents for AI
To anonymize legal documents for AI, you start from the files as they exist in the firm: written submissions in Word, exhibits and judgments in PDF, exhibit lists in Excel, letters by the dozen. Each one carries party names, addresses, a French docket number (numéro RG), a Paris bar locker number (toque), sometimes a public prosecutor or prison register number. Biffer recognizes them on your computer, with no connection, replaces them with stable aliases and highlights them in black in the output file.
On this page we go through these files one by one, how to redact them, the framework of professional secrecy and the AI uses that follow.
A lawyer's files and what they contain that is sensitive
A case file is a stack of very different documents. Here are the ones we find in every firm, with what each contains that is personal or covered by professional secrecy.
Written submissions and briefs (Word, PDF)
Party names, docket number and court in every header; in the body, the client's family, financial or medical situation, addresses, precise dates.
Writs of summons and applications (Word)
Full civil status of the parties, addresses, the bailiff's details, amounts claimed, the docket number once the case is entered on the roll.
Exhibits and exhibit lists (PDF, Excel)
Contracts, invoices, witness statements, payslips, bank statements with IBAN, police reports with report and complaint numbers, an Excel list naming every witness.
Judgments and appeal rulings (PDF)
Minute, roll and appeal numbers, names of the parties and their counsel, reasoning that restates facts from private life.
Letters and meeting notes (Word, text)
Bar numbers of sender and recipient, client confidences, emails and phone numbers, the "official" marking that changes the status of a letter between lawyers.
Criminal files and CARPA accounts (PDF, Excel)
Prosecutor, investigation and prison register numbers, hearings, expert reports, statements of client funds with a CARPA number (the French bar's client account fund).
Text PDFs, Word, Excel, CSV and plain text go through the same process (see redact a PDF); images and scanned PDFs go through built-in optical character recognition, offline, then are redacted by blacking out the pixels, with no possible restoration.
Redacting these files in practice: formats, aliases, black highlighting
Everything happens on your computer, with no connection (see offline software).
- You drop the file with the lawyer profile enabled. The local analysis highlights people, addresses, dates, IBANs, docket numbers, bar numbers and CARPA numbers.
- You untick false positives (the name of a court, an article reference) and add what is missing, all occurrences at once.
- Each item becomes a stable alias (PERSONNE_001, RG_001, TOQUE_002; the alias labels are in French), identical from one exhibit to the next for the same person.
- In Word and PDF, redacted data is highlighted in black, an option enabled by default: you see at a glance what was removed. Formatting is preserved, metadata is cleared.
- You choose whether to save the mapping table, encrypted (Argon2id and AES-256-GCM) or in plain text. With it, the real names are restored in the AI's answer; without it, the redaction is irreversible.
The queue takes a complete case file, in reviewed or automatic mode. Biffer is free for five files per month, all redaction features included, after a one-time email confirmation in the software, with no password. Beyond that, Biffer for Mac or Biffer for PC costs €10 excl. VAT (€12 incl. VAT) per computer, paid once by card via Stripe: a lifetime license for the version purchased, tied to the computer, unlimited files, each major version sold separately. Every feature is included from the free plan (see pricing).
The lawyer profile adds the procedural identifiers to the general detections. Here is the exact list from the software:
| Field | Where you find it |
|---|---|
| Toque (bar locker number) | Headers, letters between lawyers, exhibit lists |
| Numéro RG (docket number) | Submissions, judgments, notices to appear |
| Numéro de parquet (prosecutor reference) | Criminal proceedings, notices to victims |
| Minute number, roll number | Court decisions, extracts from the hearing register |
| CARPA number | Statements of client funds |
| Investigation number, appeal number | Investigation files, briefs before the Cour de cassation |
| Police report number, complaint number | Police reports, criminal complaints |
| Prison register number, case number | Prison files, internal tracking |
On top of these come the fields shared by all professions (file number, reference, contract number, employee number) and the general detections: people, emails, phone numbers, addresses, French company numbers (SIREN and SIRET), French social security number, IBAN, license plates, dates. An identifier specific to your firm can be added by keyword (see anonymization by profession).
Professional secrecy: what Article 66-5 requires
Article 66-5 of French Law No. 71-1130 of 31 December 1971 covers, in advisory work as in litigation, consultations, correspondence with the client or with other lawyers (unless marked "official"), meeting notes and every document in the file. Breach of secrecy is punished under Article 226-13 of the French Criminal Code.
Two consequences for the firm's files. Secrecy covers far more than the client's name: a docket number, a bar number, a hearing date and a court are enough to identify the case and must be redacted just the same. With offline software, your files stay in your firm from import to export. Biffer prepares, you validate: the lawyer reviews the preview, decides what stays and has the final word.
AI use cases once the files are redacted
A redacted file stays readable: PERSONNE_001 claims, PERSONNE_002 disputes, RG_001 always refers to the same case. The AI works normally, and the real people stay protected.
- Summarize forty pages of opposing submissions the day before a hearing.
- Extract from an appeal ruling the grounds accepted and rejected, with the matching passages.
- Rephrase a letter to the client in plain language.
- Draft an outline for a writ or submissions from redacted exhibits.
The redacted document can be pasted into ChatGPT, Claude or Gemini; the page anonymize before ChatGPT gives the useful settings. The answer, loaded back into Biffer with the mapping table, gets the real names back, still locally, with no connection. Biffer is designed in Lyon by Romain Rissoan, its founder, trainer and consultant, at OPCADIA.
Frequently asked questions
Is the docket number (numéro RG) really data to redact?
Yes, as soon as it identifies the case and, by extension, the parties. A docket number combined with a court and a hearing date is enough to find a file at the court registry. The secrecy of Article 66-5 covers every document in the file, well beyond the client's civil status. That is why Biffer's lawyer profile treats the docket number, the minute number and the roll number as identifiers in their own right, replaced by a stable alias (RG_001). If an analysis needs to keep them, the "keep" policy applies to that data type alone.
Can you redact a complete case file of several dozen exhibits at once?
Yes, with the queue. You drop all the exhibits (PDF, Word, Excel, CSV, text) and choose between reviewed processing, where you check the preview of each file, and automatic processing, which applies the detections to the whole batch. For a first case file, we recommend reviewed mode, long enough to confirm that your internal identifiers are recognized. Aliases stay consistent from one file to the next: the same party gets the same alias in the writ, the submissions and the exhibits, and the whole file can be given to the AI without losing the thread.
Is black highlighting mandatory in the redacted file?
It is your choice. By default, Biffer highlights each alias in black in Word and PDF documents, which makes the redaction visible on review and signals to a third party that data was removed. You can turn this option off before export to get an ordinary-looking document, where only the aliases (PERSONNE_001, RG_001) reveal the processing. Either way, the original data has left the file for good: the highlighting is a visual mark, and the alias has already replaced the text. In Excel, CSV and plain text, only the alias appears.
Your business documents anonymized, ready to be published or handed to an AI tool.
Biffer runs on your computer without sending a single file. Free for five files a month, unlimited for €10 excl. VAT per computer.
100% offline · Free: 5 files a month · No file sent