Mistral AI Now Uses User Input for Model Training by Default Across Most Tiers

Default Data Training Policy

Mistral AI has implemented a new policy where user input and output data, including conversations and documents, may be incorporated into its model training programs. This change means that data sharing for model improvement is now an opt-in feature for users, rather than requiring explicit consent.

Opt-Out Procedures for Vibe

For users of Mistral's Vibe service, opting out of data training can be done through the Admin panel. Users need to navigate to the 'Manage' section, select 'Vibe', and then disable the 'Allow your interactions to be used to train our models' toggle under the 'Privacy' section. Mobile application users (iOS and Android) can opt out by accessing 'Settings', then 'Data & Account Controls' under the 'Account' section, and deselecting the 'Enable data sharing' checkbox.

Opt-Out for Studio and API Services

Users of Mistral Studio and related API services can also opt out of data training. This is achieved by disabling a toggle under the 'Anonymous improvement data' section in the Admin panel. This action prevents API calls and associated data from being utilized for improving Mistral's services.

Implications for Users

This policy shift requires users to be aware of their data usage settings and actively manage them if they prefer not to contribute their data to Mistral's model training. The change affects various Mistral platforms, making user awareness and action critical for data privacy preferences.