I more and more think that the only way to manage online community is via invites. There are major downsides (difficulty of bootstrapping and reduced anonymity) but it gives a way to combat this. If a significant number of the users you have invited are bots you get your invite privileges revoked (or you get banned). It creates a chain of accountability and you can ban as high as necessary to severe the corrupted branch.
This feels like implementing a certificate authority system for individual users
I wonder if it is feasible to use of a web of trust that is less cumbersome and more resilient than the original GnuPG WoT, that could do the same thing. Instead of hierarchical introductions you have trusted users vouch for you not being a bot (one could even think about extending this to general rule abidance, turning it into a full on reputation system). It would feel pretty bad to loose an account, just because whoever invited you later also invited a bunch of bots/untrustworthy users