Pseudonymization software for lawyers, 100% local.
Hexagone AI is French pseudonymization software that runs entirely on the lawyer's own computer. It replaces the names of the parties, companies, addresses, amounts and references in a case file before Claude, ChatGPT, Mistral or any other AI reads it, then re-identifies the answer locally. It is the practice the French Bar's (CNB) guide on ethics and artificial intelligence asks lawyers to apply systematically.
Pleadings, contracts, the other side's exhibits, client correspondence: generative AI is useful on the whole file, and the whole file is exactly what professional secrecy covers. Hexagone AI lets you keep the AI tools you already use, on a copy where nothing identifies your client any more, without a single file leaving the firm.
What your profession's rules say about AI
- National Rules of the Bar (RIN), article 2.1
"A lawyer's professional secrecy is a matter of public policy. It is general, absolute and unlimited in time." (our translation)
No exception exists for AI tools: a prompt sent to a provider is a disclosure to a third party like any other.
- Law of 31 December 1971, article 66-5
Consultations, correspondence, meeting notes "and, more generally, every document in the file is covered by professional secrecy". (our translation)
It is not only the client's name that is protected: it is every document you might want an AI to analyse.
- CNB guide "Ethics and artificial intelligence" (2026)
"Systematically pseudonymise or anonymise data before any use of a generative AI tool." (our translation)
The guide the CNB adopted in March 2026 makes prior pseudonymisation the reference practice, and warns about indirect identification by cross-referencing data.
- CNB guide (2026), on sovereignty
"The only ways to eliminate sovereignty risks […] are: prior anonymisation or pseudonymisation of the data, or installing the LLM on the firm's own information systems." (our translation)
The guide also notes that a provider's promise not to reuse data "is generally neither provable nor verifiable". Pseudonymising on your own machine does not depend on anyone's promise.
- Paris Bar, White paper on AI (2025)
A lawyer may use anonymisation or pseudonymisation. "They may also use a locally deployed tool, with no external transmission of data to third-party servers." (our translation)
Hexagone AI ticks both boxes at once: the pseudonymisation is done by a locally deployed tool.
- French Penal Code, article 226-13
Disclosure of secret information by the person entrusted with it "is punishable by one year's imprisonment and a fine of 15,000 euros". (our translation)
The exposure is not only disciplinary.
Pasting a client file into ChatGPT, Claude or Mistral means handing it to a company that is not bound by secrecy, on infrastructure you do not control. "Don't train on my data" settings differ by plan and change over time: in 2026, for example, Mistral switched training on by default for its individual and Team plans.
Pseudonymising by hand ("Mr X", "Company Y") works for a short question. On a two-hundred-page file, with names that come back in three forms, addresses in letterheads and amounts in tables, something always slips through, and the AI loses track of who is who from one document to the next.
Online pseudonymisation tools solve the volume problem, but the original document, real identities included, goes to their server to be processed. A contract can frame that; it is not the same as transmitting nothing.
How it works
Point Hexagone AI at the case folder
PDF (scanned ones too, with local OCR), Word, Excel, PowerPoint, text. One file or a whole directory of exhibits.
Detection runs on your machine
Hexagone AI's models find identifying data and replace it with neutral markers or realistic substitutes. The same person keeps the same pseudonym in every document of the file.
You review before exporting
Every detection appears in the preview pane, and you can correct anything before exporting the protected copy. Files are rebuilt with their original layout.
Your AI works on the copy
Claude (Cowork included), ChatGPT, Mistral, Copilot: you keep your tools. They read a file where nothing identifies your client any more.
The answer is re-identified locally
The mapping never leaves your computer. Hexagone AI puts the real names back into the AI's answer, on your machine.
What gets replaced
- Names of parties, witnesses and opposing counsel, initials and aliases
- Company names, subsidiaries and trade names
- Postal addresses, emails and phone numbers
- Social security, passport and tax numbers
- IBANs, card numbers and amounts
- Dates of birth and other identifying dates
- Case numbers and client references
- Your own entity types, which you define yourself
What they say
“This partnership lets us do what many still consider incompatible: use the best AI models on the market while guaranteeing absolute confidentiality.”
“I pseudonymise every one of my documents up front with Hexagone AI. Claude NEVER touches my original files. That is the golden rule.”
Go further
Draft an NDA from your own precedents
ReadUse caseSummarize the opposing side's filings
ReadUse caseExtract and compare contract clauses
ReadGuideGenerative AI: what the French Bar (CNB) recommends to lawyers (2026 guide)
ReadGuideAnonymization or pseudonymization: what the GDPR says, and what it changes for AI
ReadGuideUsing Claude, ChatGPT or Mistral on a client file: what leaves, what stays
ReadGuideClaude Cowork for lawyers: using it without exposing professional secrecy
ReadGuideA law firm AI charter: what the Paris Bar's model contains (July 2026)
ReadFrequently asked questions
- Can a French lawyer use ChatGPT or Claude on a client file?
- The CNB guide "Ethics and artificial intelligence" (2026) says lawyers must never share data about their cases or clients with a generative AI, and that the solution is to pseudonymise or anonymise the data first. Using ChatGPT or Claude on a file pseudonymised beforehand, and systematically checking the output, is therefore the recommended practice; pasting the raw file is not.
- Is pseudonymisation enough to respect professional secrecy?
- It removes what directly identifies the client, which is the core of it. The CNB warns about indirect identification by cross-referencing (a rare job title, a small town, a date). That is why Hexagone AI also handles dates, amounts and references, why you review the copy before sending it, and why we measure re-identification risk on RAT-Bench, Imperial College London's benchmark.
- Why a local tool rather than an online platform?
- With an online platform, the original document is sent to the vendor to be pseudonymised, so you need a processing agreement and have to trust its location and retention. With Hexagone AI, the original document and the mapping stay on your computer. The Paris Bar's white paper lists a locally deployed tool, with no transmission to third-party servers, as one of the options.
- How does it differ from other pseudonymisation tools?
- Tools make different choices: some process locally, others in the browser or on the vendor's servers. Hexagone AI is software installed on the computer. Our guide to choosing pseudonymization software sets out what each architecture implies.
- Is Hexagone AI a French company?
- Yes. HEXAGONE AI is a French SAS registered in Nanterre (SIREN 989 062 849), headquartered in Boulogne-Billancourt, founded in 2025 by Mariia Zameshina and Romain Boudet. The desktop software transmits no document, so there is no data transfer to organise, in France or anywhere else.
- How much does Hexagone AI cost for a firm?
- €49 excl. VAT per user per month billed annually (€59 monthly) for Starter, €79 (€99 monthly) for Pro. One week's free trial, no credit card: you can test it on a real file, since nothing is uploaded.
Try it on a real file.
One week free, no credit card. Nothing is uploaded, so you can run it on a real file without asking anyone's permission.
macOS Apple Silicon and Windows 10/11