Guardrail classifier — AI Dictionary

A model, separate from the assistant, that screens input or output for policy violations before a response ships.