First, I wouldn't put it past OpenAI to charge even if your prompt triggered the guardrails, transferring responsibility is their style.

Second, how does vague billing prevent someone from mapping the "I'm sorry, I can't help you with that" style of responses to map the guardrails?