Mistral's own help page says input and output data, including conversations and documents, "may be included in Mistral's model training programs." Users can opt out any time, and once they do, the page says, "Mistral no longer uses your input or output data for the purpose of training its models." Only its enterprise customers get that switched off from the start, with an admin-level toggle. Everyone else starts opted in.

That's not new to the small print, but the mechanics changed for at least one paying customer. Commenter teekert says their organisation picked Mistral for its EU privacy controls, moved up to the Team tier for a dashboard that let them disable training on prompts organisation-wide, then watched Mistral change the defaults and remove that central switch. "This even caused some of our (testing) prompts to be use[d]," they wrote. Commenter floki165 calls the loss of a central opt-out "a massive red flag," arguing you can't expect every employee to remember to flip a personal privacy toggle, and that this "basically makes the Team tier unusable" for companies handling sensitive or GDPR-covered data.

Not everyone reads the page the same way. Commenter saaaaaam calls the framing "a hugely misleading editorialised title," pointing out the page itself is titled "Can I opt out of my input or output data being used for training" and leads with the opt-out right, not a training mandate. Commenter maz1b asks whether this squares with how Mistral has marketed itself as the privacy-minded European alternative.

Mistral disclosed in February that its annual recurring revenue had passed $400 million, up from $20 million a year earlier, and that it expects to clear $1 billion in ARR this year. The tier where training is off by default is the one built for the customers paying into that growth.