Mistral AI releases Shieldstral
Shieldstral released — a 3B open-weight, policy-adaptive multimodal safety classifier (Apache 2.0). It takes a natural-language policy at inference time and makes judgments without retraining
Why it matters
A design that ships a safety classifier as a small open-weight model and injects the policy through the prompt to fit each organization's standards
Source — Primary sources first; secondary reports are labelled “Press”
Officialmistral.aihttps://mistral.ai/news/shieldstral/Added in Week 35, 2026 · See that week