Mistral's newly introduced 3B Shieldstral model represents a significant advancement in AI safety protocols. By utilizing natural language yes-or-no questions, it effectively checks AI inputs and outputs for safety violations, diverging from traditional fixed category systems. Remarkably, Shieldstral performs on par with models that are seven times its size in certain benchmarks. This flexibility allows operators to define their own safety criteria in real-time, enhancing operational autonomy. Furthermore, the capability to run locally adds an additional layer of privacy and control. In this context, it is important to note that while Shieldstral offers substantial benefits, its long-term implications on industry standards and regulatory frameworks remain to be fully understood.