Mistral AI Implements Tiered Data Training Opt-Outs
AI-generated image · Bay Street Wire
New policy details reveal that while enterprise customers are protected by default, standard users must manually opt out of model training.
Mistral AI has detailed the mechanisms through which user input and output data—including documents and conversations—are integrated into its model training programs. According to documentation released by Mistral AI, the level of privacy protection depends on the specific service tier and platform utilized.
For users of Vibe, Mistral AI reports that customers are not opted out of training by default. These users must manually disable the "Allow your interactions to be used to train our models" toggle within the Admin panel or deselect the "Enable data sharing" checkbox in the iOS and Android mobile applications to prevent their data from being used. In contrast, Vibe (Enterprise) customers are opted out of training by default, with the opt-in toggle managed at the admin level.
Regarding Mistral Studio and related API services, Mistral AI states that customers can opt out of training by disabling the toggle in the "Anonymous improvement data" section of the Admin panel's privacy menu. Mistral AI notes that the opt-out toggles for Vibe and the API are separate, requiring users to configure each individually to ensure their data is not used for model improvement.