I guess it is all case by case.
This is the challenge with using a bot, because the bot is going to treat all with indifference. However as you rightfully say, for every good, there are many bad actors - so some balance must be found and the errors that take place are part of the search.
The bots definitely need to be tweaked and improved with every reasonable feedback or issue that comes up. For the most part, it saves humans a lot of boring, thankless work. If we had to do everything manually, we would lose or require many more resources.
My thought is we can look at things like HP, account age, and rep (both the number and vouching) and require manual reviews of well knwon accounts. It could be an algorithm and a whitelist could be created for those unique cases. No one would be immune to punishment or moderation, but it would prevent an automated mistake.
For sure, this is why it has to be automated to some degree at the very least - but as you said, it has to be continually tweaked to both keep up with the abuse as well as make sure it isn't doing more harm than good, the last one is hard to know.
THis is not a bad idea as I know that there are a "few" large accounts that have already been caught being pretty dodgy (to say the least) and there are quite a few of the "returned" who have immediately picked up their autovoter trails again...
No one should be immune, no matter how much stake they have :D