Shieldstral

Shieldstral is an innovative open-weight multimodal guardrail developed by Mistral, designed to enhance safety and compliance in AI-generated content. It allows users to define safety policies in natural language at inference time, making it highly flexible and user-friendly. Capable of analyzing both text and images from a single token output, Shieldstral operates locally on a modest 16GB GPU, ensuring data privacy and low latency. Its open-weight architecture means users can customize and adapt the model to specific safety requirements, making it ideal for organizations seeking robust AI safety measures without heavy infrastructure demands. Suitable for developers, AI teams, and enterprises, Shieldstral stands out by offering runtime safety customization in a lightweight package, blending cutting-edge AI with practical deployment considerations.