Earlier quoted context omitted.
You're ignoring that adding non-copyrightable code to a collection of copyrightable code makes the collection copyrightable. The folks that wrote the policy are also ignoring that. It's simply a legal misunderstanding, because (I suspect) they didn't actually consult a lawyer, and their reading of precedent is overly broad.
> You're ignoring that adding non-copyrightable code to a collection of copyrightable code makes the collection copyrightable. That is true for public domain code. It is not true (at this time) for machine generated code.
Anyway, that case did not establish that code written with an agent cannot be copyrighted by the human operating the agent. The whole problem is that the application was submitted listing the AI as sole author, so the court did not consider whether the operator could have listed himself as the author. And if he had no one would have had standing to dispute it.