Notaries choose sovereign cloud: what the CSN's announcement changes

On 8 July 2026, the Conseil supérieur du notariat (CSN) announced a partnership with Mistral AI and Scaleway to give the profession a secure, sovereign cloud environment. A few months earlier, in December 2025, the CSN had published its first AI guide for notarial offices, which already explicitly favoured French, sovereign solutions.

Taken together, these two moves say something simple: the national body of a profession bound by absolute secrecy has just settled the debate. The question is no longer whether notarial offices will use AI, or whether that AI needs to be sovereign. The question has become: what do you run on it?

What the announcement confirms

For a notarial office, this announcement removes three hesitations at once.

Legitimacy. Using AI on notarial files is no longer an isolated risk to take: it's a direction endorsed by the institution, with a framework. The notary who equips their office is following the profession's line, not stepping outside it.

The sovereignty criterion. By choosing a French model and a French host, the CSN set the bar: notarial files have no business sitting on platforms subject to foreign jurisdictions. That is exactly the case we have made from day one, and it is now institutional.

The timing. A profession-wide infrastructure isn't announced to sit empty. Offices that wait "to see" will mostly watch their peers pull ahead on a foundation that is now mapped out.

What infrastructure doesn't do on its own

It's worth being precise about what a sovereign cloud environment actually provides: a trusted place to host data and run models. That's the foundation, and it's indispensable.

But a notarial office doesn't work with a foundation. It works with files: forty documents in an estate to cross-check, a draft deed to prepare, fees and duties to calculate, a history to trace back. Between the hosted model and the finished work, an entire layer is missing:

• agents configured for the office's tasks, that read the documents and prepare the work;
• citations: every statement points back to the document it came from, because a notary's liability cannot be delegated to a machine;
• an audit trail: who asked for what, on which file, when - a record that holds up;
• governance: what the agent is allowed to read, prepare and propose, and what requires human validation.

That's the difference between having a data centre and having a colleague. The CSN's announcement installs the first. An office's daily work demands the second.

Our reading

We build exactly that working layer: sovereign agents that run on French infrastructure - the one your office chooses - and that do the repetitive work while citing their sources. The direction the profession has taken doesn't require us to change anything: it describes the architecture we already deliver, and it removes the hardest part of the conversation.

For an office, useful questions to ask right now - to us, or to any other provider:

1. Where do the models run, and under which jurisdiction, for each use?
2. Does every answer cite the document in the file it came from?
3. Is there a consultable audit trail, aligned with your professional obligations?
4. What happens if you leave: can you get your data, your configurations, your agents back?
5. Who decides what the agent can do on its own, and what waits for your validation?

A provider that answers these five questions precisely respects you. The others are selling you infrastructure with a logo on it.

If you want to see what this working layer looks like on a real file, let's talk: the demonstration runs on documents, not slides.