Llm can but do they? Is there any evidence that they spit out a piece of code verbatim without being explicitly prompted to do so? NYT v OpenAI for example, NYT intentionally prompted to circumvent OpenAi's guardrail to show NYT articles