Every AI response is shaped by built-in **guardrails**—the rules and algorithms influencing, modifying, or filtering every output. Unless the model replies with an apologetic refusal to answer, these guardrails are usually invisible and unknowable to us.