AnonymFilesMistral (Le Chat, now Vibe) and sovereign AI: is privacy enough for professional secrecy?
Mistral and privacy: Le Chat, now Vibe, hosts data in the European Union by default, but trains its models on conversations from the free, Pro and Team plans unless training is turned off; only the Enterprise plan is excluded by default. A French AI protects data better without lifting professional secrecy: French lawyers and notaries, even with the CSN's partnership with Mistral AI, only send it a client file as an anonymized copy.
Has Mistral's Le Chat become Vibe?
Mistral AI's assistant is now called Vibe. The vendor's help center announces it simply: Le Chat “is now Vibe,” its unified agent for productivity and code. Accounts, subscriptions, conversations and saved content were kept, and the address is still chat.mistral.ai. Many professionals still search for it under the name Le Chat: it is the same service, and Vibe's rules apply.
The change is not just one of name. Vibe highlights a “Work” mode that connects to e-mail, calendar and other tools. Each connector enabled widens what the assistant can read: this needs to be decided before opening it to the firm's files.
CSN and Mistral AI: what does the partnership provide for notaries?
In a news item of July 9, 2026, the Conseil supérieur du notariat (CSN, the French notaries' council) announced that it had chosen Mistral AI and the French hosting provider Scaleway to give notarial offices access to AI “in a secure, controlled European cloud environment that guarantees respect for the notary's professional secrecy” (our translation). The partnership runs for 18 months. The CSN specifies that notaries remain free to use it or not, and that they may choose other solutions that comply with the profession's AI charter.
This framework negotiated by the profession is of a different nature from an individual Vibe subscription: a notary who opens a Pro account on their own does not benefit from it. And even within the CSN environment, sending only what the task requires remains the GDPR's data minimization rule. The detail is in our article AI and notaries. To prepare these copies at the office, see our document anonymization software for notaries.
Does Mistral train its models on your data?
Mistral AI offers a free plan, a Pro plan, a Team plan and an Enterprise plan. On model training, its help center draws a clear line between individual plans and business plans.
| Plan | Training on your conversations | Who decides |
|---|---|---|
| Free, Pro, student | Yes by default: inputs and outputs are used to train the models | The user, who can opt out at any time in the settings (privacy) |
| Team | Yes by default, unless the administrator turns it off | The administrator, who can turn it off for the whole organization |
| Enterprise | No by default | The administrator, who manages any activation |
| All | A thumbs-up or thumbs-down rating authorizes use of the rating, the input and the output | The user, with each rating |
Two reflexes follow. On an individual plan, turn off training before any professional use, on the web as well as in the mobile app. On every plan, never rate a work conversation: the rating counts as authorization, even when training is turned off. Finally, Mistral's API has its own setting: turning off one does not turn off the other.
Where does Mistral host data? What is guaranteed and what is not
Mistral AI states that “by default” data is hosted in the European Union. Its help center adds three nuances:
- a customer who explicitly uses the API's US endpoint has their data hosted in the United States;
- depending on the feature used, data may be temporarily transferred outside the Union, to the subprocessors listed in Mistral's Trust Center, with the safeguards of Article 46 of the GDPR (the European Commission's standard contractual clauses);
- Enterprise customers can have certain features involving these transfers turned off for their organization.
This is a real advantage from a GDPR standpoint. But “hosted in Europe by default” is not “never outside Europe”: sovereignty must be checked along the whole chain, subprocessors and connectors included. This is also what matters with regard to the US Cloud Act, which allows US authorities to request data from a provider subject to their jurisdiction, wherever that data is stored.
Does a sovereign AI exempt a lawyer from anonymizing?
Sovereignty answers the question “who could access this data, and under what law?” Professional secrecy asks a simpler question: “have I disclosed the information to anyone?” Article 226-13 of the French Criminal Code (Code pénal) punishes the disclosure of secret information “with one year of imprisonment and a fine of 15,000 euros” (our translation), without distinguishing by the recipient's nationality or the location of its servers.
Handing a raw document to Vibe means handing it to Mistral AI and its subprocessors. A French AI, hosted in France, under a processing agreement, reduces the risks: it does not make the disclosure disappear. This is why the professions' guides, such as that of the Conseil national des barreaux (French national bar council), ask that no file or client data be put into it, whatever the tool. Our guide AI and professional secrecy brings together the rules profession by profession. For a law firm, see our document anonymization software for lawyers.
How do you use Mistral on a client file?
- Set up the account. Prefer a Team or Enterprise plan administered by the firm. On an individual plan, turn off training in the privacy settings, on the web and on mobile. Do not enable any connector to the firm's e-mail or files without having decided to.
- Prepare the copy on the computer. With AnonymFiles, drop in the file: PDFs, scans read through local OCR, Word, .eml and .msg e-mails, Excel, CSV, PowerPoint. Names, organizations, addresses, e-mails, phone numbers, IBANs, private dates and ID numbers are replaced with stable labels such as [PERSONNE_1].
- Review side by side. Each original is shown next to its copy; one click masks whatever the detection missed. Nothing is exported without approval.
- Export. A Word file becomes a .docx redacted in place, layout preserved; a PDF is rebuilt with the content truly deleted. The mapping table is never part of the export.
- Work with Vibe. Paste the instruction provided by the app so that the AI keeps the labels intact, attach the copy, write your request. Do not rate the conversation.
- Re-inject. Bring the answer back into the app: it puts the real names back as text, Word or PDF, on the computer.
On the network side, the app works with no account and no telemetry, and its license is checked on the computer, with no network call. On Mac, the detection models are included; on Windows, they are downloaded once at the first processing run, without any document text. The security page details what the app does and does not do on the network. Pricing: €290 excl. VAT per seat per year for one or two seats, €240 excl. VAT per seat from three seats, quote above twenty seats.
Frequently asked questions
Can a lawyer use Mistral (Le Chat, Vibe)?
Yes, to draft, summarize or rephrase, on a copy in which the client can no longer be recognized. The guide of the Conseil national des barreaux (the French national bar council) asks lawyers not to send any file or client data to a generative AI without prior pseudonymization or anonymization, whatever the vendor. Prefer a Team or Enterprise plan managed by the firm, and turn off training.
Is Mistral GDPR compliant?
Mistral AI hosts data in the European Union by default and publishes a data processing addendum (DPA). Compliance still depends on how you use it: legal basis, data minimization, informing the people concerned. And the GDPR does not settle professional secrecy: for a client file, the anonymized copy remains the rule.
Does Mistral's Le Chat still exist?
Le Chat became Vibe in 2026. According to Mistral AI's help center, accounts, subscriptions, conversations and saved content were kept, at the same address: chat.mistral.ai. The privacy rules described here apply to Vibe.
Does Mistral train its models on my conversations?
On individual plans (free, Pro, student), yes by default: inputs and outputs are used to train the models, unless you turn off the option in the settings. On the Team plan, the administrator can turn it off for the whole organization. Enterprise customers are excluded from training by default. A thumbs-up or thumbs-down rating authorizes use of the conversation, whatever the plan.
Where does Mistral host data?
In the European Union by default, according to Mistral AI's help center. Some features may, however, temporarily transfer data outside the Union, to the subprocessors listed in its Trust Center, with the safeguards of Article 46 of the GDPR. Enterprise customers can have some of these features turned off.
Does a French or EU-hosted AI respect professional secrecy?
Not on its own. Professional secrecy forbids disclosing information to a third party, whatever its nationality or the location of its servers. Hosting in Europe settles part of the question of the GDPR and foreign laws; it does not make the AI vendor the client's confidant. A document handed over without being anonymized is still disclosed.
Do notaries have to use the CSN's Mistral solution?
No. On July 9, 2026, the Conseil supérieur du notariat announced an 18-month partnership with Mistral AI, on the infrastructure of the French hosting provider Scaleway, and specifies that notaries remain free to use it or not, and to choose other solutions that comply with the profession's AI charter.
How do you use Mistral on a client file?
Turn off training, do not send feedback on a work conversation, and above all give the AI only an anonymized copy prepared on the computer: names, addresses, amounts and identifiers replaced with labels such as [PERSONNE_1]. The real names are then put back into the answer, on the computer.