That's a lot of words to argue that we shouldn't be separating prompts from the (often untrusted) data the prompts are acting on.

I still don't see your point.