There is no reason to think these are the actual instructions. LLMs simply try to produce statistically likely content based on your prompt.