This analysis describes what Mistral AI's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
The provision establishes the operational basis for Mistral AI's model training methodology and defines the sources from which training data is sourced. It establishes that personal data filtering occurs at multiple levels (both by third parties and by Mistral AI) but does not guarantee complete removal of personal information from training datasets.
The updated policy now explicitly includes data accessed through third-party services and integrations users connect to Mistral AI Products as 'Input' data subject to collection and use. The policy removed its prior statement that Input and Output data are not used to train AI models when using Le Chat Enterprise or paid versions of Mistral APIs. This creates operational ambiguity: users of paid services and Enterprise customers no longer have a documented commitment that their data will be excluded from model training, though the privacy policy does not affirmatively state that model training now occurs. The policy also changed language describing product improvement from 'aggregated and anonymous statistics' to 'aggregated or anonymous datasets or statistics,' broadening the stated scope of what can be collected for improvement purposes.
View change record →Under this clause, individuals whose personal data appears in publicly available Internet sources or third-party datasets may have that data included in Mistral AI's model training processes, subject to filtration practices that are described as good faith efforts but not as absolute safeguards. Users are informed that personal data may remain present in training datasets despite these filtering practices.
How other platforms handle this
We generate new information from other data we collect to derive likely preferences or other characteristics. For instance, we infer your general geographic location based on your IP address.
To achieve these processing purposes, we use algorithms to recognize patterns in Service Data, manual review of Service Data (such as when you interact directly with our billing or support teams)...
We may analyze personal data we have collected about you to create a profile of your interests and preferences so that we can contact you with information that is relevant to you.
"Data publicly available on the Internet. Our artificial intelligence models are trained on data that is publicly available on the Internet by third parties, which may contain personal data, even if we use good practices to filter out such personal data. [...] Training Datasets. In some cases, we access datasets provided by third parties for our model training purposes. These datasets may include personal data (even if such third parties and Mistral AI use good practices to filter out such personal data), proprietary data, or public data.Excerpt from Mistral AI's Privacy Policy
How Meta, TikTok, and Supabase restructured governance language across documents, jurisdictions, and consent frameworks through incremental document updates.
How 10 AI platforms describe the use of user data for model training, improvement, and development, based on archived governance provisions.
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
The provision establishes the operational basis for Mistral AI's model training methodology and defines the sources from which training data is sourced. It establishes that personal data filtering occurs at multiple levels (both by third parties and by Mistral AI) but does not guarantee complete removal of personal information from training datasets.
Under this clause, individuals whose personal data appears in publicly available Internet sources or third-party datasets may have that data included in Mistral AI's model training processes, subject to filtration practices that are described as good faith efforts but not as absolute safeguards. Users are informed that personal data may remain present in training datasets despite these filtering practices.
ConductAtlas has identified this type of provision across 217 platforms. See the full comparison.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Mistral AI.