Mistral AI Data Training Opt-Out Policies

Mistral AI has clarified its data governance policies, confirming that user input and output data—including conversations, uploaded documents, and other user-provided content—may be used to train its models. While enterprise customers are opted out of this training by default, standard users must manually opt out through their account settings.

Training Defaults by Service and Plan

Mistral's training policies differ based on the platform and the subscription tier. The primary takeaway is that training is enabled by default for most consumers, while the Enterprise tier provides a privacy-first default.

Vibe (Consumer and Team Plans)

For users of Vibe, training is enabled by default. Users must manually navigate to their settings to disable data sharing.

Vibe (Enterprise Plan)

Enterprise customers are opted out of training by default. The ability to opt in or out is managed centrally at the administrator level.

Mistral Studio and API

Users of Mistral Studio and related API services can opt out of data training via the Admin panel. The documentation does not explicitly state the default status for these services, though they are managed via a separate toggle from the Vibe settings.

How to Opt Out of Data Training

Because Mistral uses separate toggles for different services, users must configure each one individually to ensure full privacy.

Opting Out of Vibe (Web Admin Panel)

1. Access the Admin panel.
2. Select Vibe under the Manage section.
3. In the Privacy section, disable the toggle labeled "Allow your interactions to be used to train our models".

Opting Out of Vibe (Mobile App)

1. Open the application Settings.
2. Select Data & Account Controls under the Account section.
3. Deselect the "Enable data sharing" checkbox.

Opting Out of Mistral Studio and API

1. Access the Admin panel.
2. Open the Privacy menu in the left-hand navigation bar.
3. Under the "Anonymous improvement data" section, disable the toggle to prevent API calls and related data from being used for service improvement.

Community Insights and Privacy Concerns

Following the disclosure of these policies, technical users and privacy advocates have raised several concerns regarding the implementation of these controls.

Centralized Control for Team Plans

Some users have noted a critical gap in the Team plan. Unlike the Enterprise tier, the Team plan reportedly lacks the ability for an organization administrator to centrally enforce an opt-out for all members. This means individual employees must manually disable training in their own settings, which increases the risk of sensitive corporate IP being used for training.

"Taking away the ability to centrally enforce an opt-out across an organization is a massive red flag... For any company dealing with sensitive IP or GDPR-compliant data, this basically makes the Team tier unusable."

Comparison with Competitors

Users have compared Mistral's approach to other frontier AI labs. While some argue that most AI companies follow similar patterns of data collection, others point to competitors like Claude, which reportedly disables training on prompts for organization plans starting at lower price points.

Legal and GDPR Implications

There is ongoing discussion regarding the PII (Personally Identifiable Information) handling within these prompts. Some users question whether a general warning to "not enter PII" is sufficient under European law (GDPR) if the data is subsequently stored and used for training weights.

The "Training" Definition

Critics have pointed out that "not training on your data" may be a narrow technical definition. They suggest that companies might still use data for "market research," "service improvement," or "analysis" via separate embedding models, even if the data is not used to directly update the weights of the primary frontier model.