AI Pulse · Frontier Labs Watch

Mistral has changed the default privacy setting for most Le Chat users. Input and output from Free, Pro, Student, and Team accounts can now be used for training unless the user opts out. Enterprise is the exception, with training turned off by default.

For a business owner, this is not a small settings change. It means someone on a paid Team plan could paste customer notes, internal plans, pricing discussions, support tickets, or draft documents into the tool without realising those conversations are included in Mistral’s training policy by default. The practical question is simple, do your people know which AI tools they use, what data they put into them, and what each provider’s default terms allow?

The timing has drawn attention because Anthropic reportedly moved in the other direction this week after criticism over a similar data-retention issue. Mistral’s policy is documented in its own help centre, and the change is already prompting a busy discussion among technical users. The lesson is broader than one vendor. Consumer and team AI plans are not the same as enterprise data controls, even when they sit under the same product name. This is the kind of thing we build into an AI command centre, a clear view of approved tools, data rules, and settings that need active review.