Bingo. If at the limit, the purpose of an AI is to be indistinguishable from human generated content, then watermarking and AI detection algorithms will be absolutely useless.

The ONLY means to do this is to have creators verify their human-generated (or vetted) content at the time of publication (providing positive proof), as opposed to attempting to retroactively trying to determine if content was generated by a human (proving a negative).