Mistral AI announced Shieldstral, a new 3B open-weights model for content safety.

> A 3B open-weight, policy-adaptive multimodal safety classifier that matches models up to 7x its size on text safety and sets a new state of the art on multimodal moderation.