Mistral has announced that under certain conditions, user input and output data may be included in the company's model training programs. This includes conversation content, documents, and other content provided by users.
Mistral Changes Policy to Use User Input for Model Training by Default
This article is a translation. Read the Japanese original
Customers on the enterprise tier are excluded from training by default, with administrators controlling the opt-in toggle. For all other users, the setting is configured to use data for training by default.
Users can opt out of training at any time through their settings. This can be done by disabling the "Allow your interactions to be used to train our models" toggle in the Vibe control panel. In the iOS and Android apps, users can opt out by unchecking "Enable data sharing" under "Data & Account Controls" in the settings.
Furthermore, opt-out settings for Mistral Studio and API services must be managed separately from Vibe settings. For the API, settings are configured in the "Anonymous improvement data" section within the privacy menu of the control panel. Once the setting is confirmed, Mistral will no longer use input and output data for model training purposes.
Source: Mistral now trains on user input by default, except on enterprise tier (Hacker News Frontpage, 2026-09-02)