Rules defining prohibited behaviors for an AI system, implemented through training, prompts, and filtering. Safety policy affects AUP enforcement, duty-of-care arguments, and user expectations; it is often documented and traceable to implemented controls.
See: AUP; Guardrails; Moderation