Mistral AI has developed Shieldstral 1.0 3B, a content moderation tool that uses a single yes/no question to assess safety instead of a fixed harm taxonomy. This approach allows operators to supply a policy query at inference time, receiving a calibrated safety score without requiring retraining. Built on a 3.3 billion parameter base model, Shieldstral 1.0 3B has demonstrated high accuracy in text and multimodal safety assessments, while maintaining a relatively small memory footprint. The tool is available under an open-source license. This development could improve the efficiency and effectiveness of content moderation processes.