Mistral open-sources Large 4 weights despite cyberattack uplift warning
Original title: Mistral open-sources AI model despite warning it could aid cyberattacks
Downloaded weights sit outside API guardrail updates, and Mistral has not said whether Large 4 meets the EU systemic-risk threshold.
Mistral AI released open weights for Mistral Large 4, called “Le Chonk,” after a public preview in which it monitored cybersecurity use. The company said the model refused black-box attacks on live targets but saw potential uplift for real attacks, and it is changing guardrails that reach only API and hosted users. Hugging Face noted that open weights leave policy and deployment to the user. Mistral has not published the technical report behind its coding-benchmark claims or said whether the model exceeds the EU AI Act’s 10^25 FLOP presumption.
Unconfirmed: Mistral has not said whether Large 4 is a systemic-risk model or how guardrail updates apply to weights already downloaded.
- Large 4 refused 96.6% of malicious prompts on named cyber benchmarks, Mistral says.
- It scored 51% on the Artificial Analysis Cyber Index, below two open-weight models shown.
- A vetted penetration-testing variant is in preview for partners such as Orange Cyberdefense.
- API pricing is $0.70 per million input tokens and $2.10 per million output tokens.
- A modified MIT license covers use and fine-tuning; model-as-a-service needs a separate license.
- Vendors
- Mistral AI
- AI models
- Mistral Large 4
From the original article
4 min read at csoonline.com
A trillion-parameter AI model that its developer says could help attackers carry out real cyberattacks is now free to download, modify and run outside the developer’s control. Mistral AI released the open weights of Mistral Large 4, which it calls “Le Chonk,” on Thursday, weeks after putting the multimodal model into public preview.
csoonline.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.