Rules or filters that limit what an AI can say or do, usually for safety or compliance.
"Guardrails prevented the bot from giving financial advice outside approved guidelines."
