This opens up entire conversations that can't be compressed in just a post, but I'd say that as AI models become more sophisticated, they can be utilized for identifying vulnerabilities in cryptographic domains. I'm not entirely sure about the considerations that need to be taken, but I know that we can't rely exclusively on a creation that holds no responsibility for what it produces. (That under the hypothesis that we can't blame the AI developers for an AI false output)
Likely not, and not for anywhere in the foreseeable future. GPTs are good at making up stuff with prompts but they aren't good at math. If they were to come up with mathematical theory and proofs, then I would be convinced but I don't think we are anywhere near that yet.
A public ledger can also cause public harm. Consider when governments go rogue--they will use public ledgers to track and punish enemies and dissidents.
Another scenario is medical and financial information that is required by law to have protection from public view.
There are many legitimate reasons to have transaction privacy.
Bitcoin provided sufficient privacy while keeping it simple and easy to understand. If used correctly with the appropriate tools (mixers, coinjoins, etc), then it would provide very good privacy where adversary would have a hard time trying to prove any plausible link.
The issue with privacy doesn't actually lie within Bitcoin, but it is mainly about the user's behaviors. Most privacy leakage are caused by the user's privacy habits that doesn't lie with Bitcoin.