MistralAI-Training
Mistral AI · Model training · robots.txt token mistralai-training
What it is definition
Crawls web content to build datasets for training Mistral's generative AI models.
MistralAI-Training is operated by Mistral AI and classified here as model training. These agents collect text and images that may end up in a training corpus. Disallowing one withholds new content from the next model. It does not withdraw anything collected before the disallow, and it does not remove the site from any answer engine.
Disallowing mistralai-training withholds content crawled from that point on from Mistral AI's training data. It does not withdraw pages collected in earlier crawls, it does not reach copies already sitting in third-party datasets, and it does not remove the site from any answer surface.
- robots.txt token
- mistralai-training
- Operator
- Mistral AI
- Purpose
- Model training
- Honors robots.txt
- yes
- Documentation
- https://docs.mistral.ai/robots/
Identity on the wire exact string
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-Training/1.0; +https://docs.mistral.ai/robots)curl -A 'Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; MistralAI-Training/1.0; +https://docs.mistral.ai/robots)' https://example.com/Replace the host with your own. A 403, a challenge page or an empty body means the edge is blocking the agent whatever robots.txt says.
How to allow it, how to block it robots.txt
User-agent: mistralai-training
Allow: /Absent any rule, the agent is already allowed, so this group only matters when your * group disallows something. Naming the agent replaces the * group for it entirely, so repeat here every disallow you still want to apply to it.
User-agent: mistralai-training
Disallow: /Disallowing mistralai-training withholds content crawled from that point on from Mistral AI's training data. It does not withdraw pages collected in earlier crawls, it does not reach copies already sitting in third-party datasets, and it does not remove the site from any answer surface.
Sites that block it 10 shown
| Domain | AI access score | Rank |
|---|---|---|
| msn.com | 34 | 62 |
| techradar.com | 83 | 2,215 |
| livescience.com | 81 | 2,585 |
| flipboard.com | 68 | 3,286 |
| tomshardware.com | 83 | 3,331 |
| space.com | 79 | 3,873 |
| tomsguide.com | 78 | 4,787 |
| pcgamer.com | 81 | 5,026 |
| gamesradar.com | 79 | 6,377 |
| theweek.com | 78 | 8,866 |
Compare Mistral AI
Mistral AI runs 3 agents in this registry, and they do different jobs. Blocking one says nothing about the others.
Common questions answered
Does blocking MistralAI-Training remove me from Mistral AI's AI products?
No, not by itself. MistralAI-Training collects content for training, not for the index that answers questions. Mistral AI's answer surfaces are governed separately by mistralai-user and mistralai-index, which stay allowed unless you disallow them too. Content collected before the disallow is not withdrawn.
How do I verify MistralAI-Training requests are genuine?
Mistral AI publishes neither reverse DNS names nor address ranges for MistralAI-Training, so a request carrying this user agent cannot be authenticated. The string is trivially spoofed and is routinely reused by unrelated scrapers. Treat it as a hint, rate-limit by source address, and do not grant it access you would not grant an anonymous client.
Check your own site free
See whether your robots.txt admits MistralAI-Training today, and whether your edge agrees with it.