This is just an idea, but I see that some instance prevent bots from bypassing the registration step by asking unexpected questions like “are you human?”. Apparently this is the type of question that can easily discern a human answer from a chatGPT answer. So what if the admin of an instance changed the question asked for registration application every few days? Something like " What did you hit for break f4st? " would give bot proof applications and the answers would likely give the admin a good laugh. Not 100% effective of course, but would help for the time being.
https://xkcd.com/810 may be relevant here. Mind you, that’s from 2010, long before ChatGPT.