It's only hard if your business model is "growth & engagement" and you need to maximize user and engagement numbers at all costs.
If that's not your business model, abuse prevention is trivial. You can operate the network like a members' club where people gain privileges (such as posting links, media, etc - anything that can be used to spam or harm other users/the platform) over time as they prove themselves and acquire trust (Stack Overflow calls this number "reputation") and you can then use this trust number as a weight in automated decisions, so that high-trust users (who are unlikely to suddenly burn their hard-earned account) will not be impacted by an automated ban.
Forums in the good old days were ran by volunteers were able to deal with spam/abuse just fine with a combination of bans and privilege levels (it will take time & effort to level up an account to where it's able to post links/etc and be useful for spamming), there's absolutely no reason current social media companies can't do the same, if it wasn't for the fact that their business model to a certain extent relies on moderation being both unfair to users and subpar at effectively suppressing bad content (hint: bad content is nice to have around as long as it's not too visible, as it generates tons of outrage and thus engagement - it's only a problem when powerful people get wind of it and then you delete it and issue a fake apology).
StackOverflow and "forums in the good old days" are 1- way smaller than any of Twitter/FB/etc and 2- are mostly centered on one subject: it’s quite easy to remove abuse on SO because it’s about coding and anything else should be removed. On the other hand, Twitter/FB/etc have hundreds of millions of users posting all sorts of things. Moderation is a hard problem because it doesn’t scale.
There are ~21M questions on StackOverflow, which is equivalent to the number of tweets posted every a single hour and the number of items shared on Facebook every 8 minutes.
If that's not your business model, abuse prevention is trivial. You can operate the network like a members' club where people gain privileges (such as posting links, media, etc - anything that can be used to spam or harm other users/the platform) over time as they prove themselves and acquire trust (Stack Overflow calls this number "reputation") and you can then use this trust number as a weight in automated decisions, so that high-trust users (who are unlikely to suddenly burn their hard-earned account) will not be impacted by an automated ban.
Forums in the good old days were ran by volunteers were able to deal with spam/abuse just fine with a combination of bans and privilege levels (it will take time & effort to level up an account to where it's able to post links/etc and be useful for spamming), there's absolutely no reason current social media companies can't do the same, if it wasn't for the fact that their business model to a certain extent relies on moderation being both unfair to users and subpar at effectively suppressing bad content (hint: bad content is nice to have around as long as it's not too visible, as it generates tons of outrage and thus engagement - it's only a problem when powerful people get wind of it and then you delete it and issue a fake apology).