I also think OpenAI will improve their models to be undetectable. Even now, a proper prompt might pass all those online GPT checkers. Personally, I'd like to get access to the one used by Google, but of course that's an impossible dream.
Anyway, a simple built-in synchronizer allows anyone to cheat any online checker with ease. DeepL translation of a GPT-generated text and/or the old text spinners from 2010's also work just perfectly. All these methods can be used in a fully automated way, so autoblogging is alive again and now I don't see anything that can stop it.
Human blog writers and obsolete. Period.
