Content moderation has always been a square-peg-round-hole problem. You ship a model, bolt on a guardrail trained on some fixed taxonomy of harm categories, and then discover that what counts as "unsafe" on a mental-health platform is completely different from what counts as "unsafe" on a cybersecurity research tool. Retraining for every new deployment context is expensive and slow. Mistral's answer is Shieldstral, a 3B open-weights safety classifier that sidesteps the taxonomy problem entirely by letting you write the policy yourself, at inference time, in plain English.

One question to rule them all

Shieldstral formulates content moderation as a binary question-answering task, unifying diverse moderation tasks into a single yes/no problem and enabling heterogeneous safety datasets with divergent taxonomies to be consolidated under one training framework. In practice, every moderation call is structured as three parts:

• Instruction , sets the evaluation context, strictness level, and optionally defines what "unsafe" means for your deployment.
• Query , a plain-language yes/no question, e.g. "Does this content promote physical violence?"
• Document , the content to judge: a user prompt, a model response, a prompt-response pair, or an image with optional text.

Rather than outputting fixed categories, Shieldstral takes a natural-language query describing a safety concern and a piece of content to evaluate, and produces a single continuous safety score. That score comes from reading only the yes and no logits from the final token and softmax-normalizing them , a single forward pass, no chain-of-thought overhead.

This framing is deceptively powerful. The same checkpoint handles prompt classification, response moderation, refusal detection, and toxicity detection. Swap the query, and you've retargeted the model to a completely different policy without touching the weights.