R to @MistralAI: Shieldstral is available under Apache 2.0. Try it…

TL;DR - Mistral AI announced that Shieldstral, a 3B-parameter safety/guardrail model, is released under the Apache 2.0 license and available on Hugging Face. It matters because permissively licensed moderation models let teams self-host content-safety filtering without depending on proprietary APIs.

• Company announcement from @MistralAI (official account), not an individual take — a product/model release rather than commentary.
• Model identifier is mistralai/Shieldstral-1.0-3B, indicating a compact ~3B-parameter model sized for cheap, low-latency guardrail inference alongside larger LLMs.
• Apache 2.0 licensing permits commercial use, modification, and redistribution, aligning with Mistral's open-weights strategy.
• Content is thin: the post gives only the license, name, and Hugging Face link — no benchmarks, taxonomy details, training data, or evaluation results are provided.