Sure, but LLMs are inherently lossy. There is no guaranteed way for the second AI to extract the original prompt from the text.

It's almost as though someone was put in charge of AI growth and all they care about is token burn.