232 APPS TRACKED · 227 CLAUSES ON FILE · 38 WITH NO CLAUSE TO QUOTE
RECORD OPENED 5 OCT 2026 · 2 CLAUSES ON FILE
ai-assistants

Does Mistral AI train on your data?

Mistral does not train on your data "except (a) when you have not opted out" — a double negative meaning it trains by default. The switch is real, named, and in your account settings.

LAST CHECKED
5 Oct 2026 · 5 days ago
CONFIDENCE
high
RIGHT OF REPLY
Not yet asked
Verdict
C
Trains on you by default
A self-serve opt-out exists

Evidence 1 — Consumer accounts — European Union
Terms of Service · Consumer accounts — European Union
EVIDENCE 01 OF 02
"We do not use Your Data to train our artificial intelligence models except (a) when you have not opted out of training or (b) when you provide Feedback to us."
CHECKED
5 Sept 2026
STILL LIVE AS OF
5 Oct 2026
APPLIES TO
Consumer accounts — European Union
JURISDICTION
eu
HOW IT WAS TAKEN
Fetched from the company
SHA-256
A5FF0A11…3B7F

How we read it: Read the clause twice: a denial wrapped around an admission, and what it admits is that training runs until you turn it off. Credit where due — Mistral states its default, which is the one thing the eight companies grouped on /patterns never do, and the privacy policy says where the control lives: "directly from your account". That is what makes this a C rather than a D. The hole is limb (b). A thumbs-up or thumbs-down sends that Input and Output to training whatever your setting says, and no opt-out reaches it. Here the carve-out is at least bounded, "to the extent strictly necessary"; in the rest-of-world terms it is not.

CONTROLS SEARCHED FOR · 29 Aug 2026

opt out, opt-out, object, user control, settings, withdraw

What the document does offer

  • "We've introduced a user control which allows you to object to the use of your input and output data for model training directly from your account" — named, located, and tied to training rather than to advertising
  • "user controls which are available from your Mistral AI Account settings and which allow you to: delete your account at all times, obtain a copy of your data, object to the use of your input and output data for model training"
  • "when your request concerns the training of our models, it's important to note that your rights have technical limitations" — a caveat on removal from a model already trained, not on the switch itself

Why none of it applies: The control is named, its location is stated, and the default it changes is stated too, which together clear the bar the D-grade companies miss. What the documents do not do is bring Feedback inside it.

Evidence 2 — Consumer accounts — rest of world
Terms of Service · Consumer accounts — rest of world
EVIDENCE 02 OF 02
"We do not use Your Data to train our artificial intelligence models except (a) when you have not opted-out of training or (b) when you provide Feedback to us."
CHECKED
1 Oct 2026
STILL LIVE AS OF
5 Oct 2026
APPLIES TO
Consumer accounts — rest of world
JURISDICTION
global
HOW IT WAS TAKEN
Fetched, browser-rendered
SHA-256
1A38E84B…30C3

How we read it: The same clause and the same default as the EU tier: training is on until you turn it off. What differs is not the hyphen in "opted-out". Where the EU terms bound the Feedback carve-out to "the extent strictly necessary", this document does not — it reads "Mistral AI may use your Feedback, along with the associated Input and Output data, for the purpose of training our artificial intelligence models and improving the Mistral AI Products", and stops. So a reader outside the EU who rates an answer hands over that exchange on unbounded terms, and the opt-out does not reach it. Two points worse for that, on the same grade: the difference is the size of a hole, not whether there is one.

What each plan can refuse

The same questions, asked of every plan we could identify. Every cell comes from a quoted clause or a documented search - see the evidence above.

Consumer accounts — European UnionTRAINS: YES
OPT-OUT
Self-serve toggle
RETENTION
Not stated
EXPOSURE
38 / 100
DE-IDENTIFIED
Not stated
HUMAN REVIEW
Unknown
Consumer accounts — rest of worldTRAINS: YES
OPT-OUT
Self-serve toggle
RETENTION
Not stated
EXPOSURE
38 / 100
DE-IDENTIFIED
Not stated
HUMAN REVIEW
Unknown

How to opt out of Mistral AI training

Plan by plan, from what Mistral AI's own policy says about refusing. Where it publishes the steps, they're listed here with the limits they carry. Where there's nothing to switch off, or nothing graded, this says so instead of inventing a menu path.

Consumer accounts — European Union

Yes. Mistral AI has a self-serve setting.

Le Chat (now Vibe), web and app
  1. Open Settings.
  2. Under Account, select Data & Account Controls.
  3. Untick "Enable data sharing".

Applies from the point you opt out; Mistral's terms say removal from a model already trained has technical limits.

Consumer accounts — rest of world

Yes. Mistral AI has a self-serve setting.

Le Chat (now Vibe), web and app
  1. Open Settings.
  2. Under Account, select Data & Account Controls.
  3. Untick "Enable data sharing".

Applies from the point you opt out; Mistral's terms say removal from a model already trained has technical limits.

Each step's source and archived copy →

What this policy has done since we started watching

9 CAPTURES SINCE 29 AUG 2026

3 changes detected since 29 Aug 2026. Each one is a diff between two archived copies, and each links to the passage that moved.

  1. NotableTouched opt-outs, retention, what's collected, sub-processors and plan tiers.
  2. MinorTouched opt-outs. Not yet reviewed by a human.
  3. MajorTouched opt-outs, retention, what's collected, sub-processors, plan tiers and training.

Everything we read

DOCUMENT
CHECKED
SNAPSHOT
AI Addendum
5 Sept 2026
Privacy Policy
14 Sept 2026
Terms of Servicequoted
5 Sept 2026
Terms of Servicequoted
1 Oct 2026

Vendor response

From the company? What a reply does and what moves a grade.

We haven't yet written to Mistral AI about this verdict, and they haven't disputed, clarified or announced a change to any clause quoted above. When either happens it's published here, word for word.

Report a wrong clause

Compared with OpenAI and Anthropic · Writing about Mistral AI? Embed this verdict as a card that stays current.