The moderation is mostly federated. You can choose which lists to subscribe to to mute/block/label as you see fit.
From what I can tell, most content moderation on bluesky is done via labelers, which you need a pds to run, but not to use.
Both X and bluesky have semi-automated moderation processes with an inherent false positive rate. The two high profile cases I'm aware of (Jessie Singal and Laura Loomer) that were suspended automatically eventually got turned back on once a human was involved.
https://bsky.app/profile/jessesingal.com
https://bsky.app/profile/loomered.bsky.social/post/3lemtdvls...