It’s a Lemmy feature. Every instance can have a list of slurs that’s automatically removed from all messages. You can see the instance’s slur regex in the /site api endpoint, key slur_filter_regex. Lemmy.ca’s filter bans the word “removed” among some other things.

I’ve looked at a few other instances and they’re all interestingly different. Someone should do data science on this. E.g., I’ve yet to find an instance that uses it to automatically censor ideologically opposing sites, which is better than I expected, but it’s almost certain that some instances do.