Mistral now trains on your prompts by default on Free, Pro and Team
The org-wide opt-out is gone from the Team plan, now Enterprise-only. Here's how to switch training off per account, and how Claude and Grok defaults compare.
Copy markdown
Default-on, and it slipped past the docs
On Free, Pro and standard Vibe/Studio accounts, “use my data for training” is now enabled by default — and the switch reportedly contradicted Mistral's own documentation for several days. If you've prompted Mistral on those tiers lately, assume those inputs and outputs were eligible for training unless you'd already opted out.
The Team org-wide toggle is now Enterprise-only
The single organization-level switch that let a Team admin opt everyone out has reportedly been removed and reserved for Enterprise. On Team, each member now has to disable training on their own account — there's no way left to enforce it across a whole org below the Enterprise tier.
How to actually opt out today
In Vibe, turn off “Allow your interactions to be used to train our models” under Privacy (on mobile: Settings → Data & Account Controls → deselect data sharing). The API/Studio has a separate switch under Privacy → Anonymous improvement data, and the two don't sync — so flip both. Note that thumbs-up/down feedback is always used, whatever your toggles say.
How Claude and Grok compare
Per the thread, Anthropic keeps prompt-training off for organizations from its ~€18 Team tier upward, and xAI's Grok defaults training off even on the free tier. If org-level data control is a hard requirement and you're on Mistral Team, that guarantee now sits behind an Enterprise contract.