Safety Guardrails

Rules and filters applied around a model at the moment of use, checking inputs and outputs to block harmful, illegal or policy-breaking content. They complement model alignment rather than replace it: guardrails control what goes in and out, while alignment shapes the model's own behaviour.