Guardrails are mechanisms to guide the responsible development of AI so that it does not become a technological, social, or security threat.