this is a great point, thank you for posting it.
i've been researching this problem for a year now, and came across some interesting technical and social povs on moderation and network attacks.
- moderation: my pov is that large companies with teams of moderators (such as twitter or facebook) wont and cant scale to varied moderation policies across the world. so the task of moderation has to be boiled down to the communities themselves. the question then is - a) how do you prevent a moderator from getting power hungry, b) how do you incentivize them to do the work and c) how do you make this auditable and provide recourse mechanisms? for all of this, open access to an API to build the right tools is super important as you said. and having this done on a protocol is the only real way to scale it imo.
- network attacks: we've done a lot of work in approximate solutions for sybil attacks. using verifiable delay functions (VDFs) as a way to order messages, and having them stored on multiple locations (eg arweave) in append only ways gives a lot more resilience / provides a way towards solutions here.