Take the use of reclaimed slurs, for example. When used against the discriminated group by a dominant group, their intention is often to cause harm. When used within the group, the intention is to reappropriate the term: https://en.wikipedia.org/wiki/Reappropriation
Similarly, harassers will use terms in ways that are plausibly read different ways depending on who they're talking to. So something that might sound innocuous or just odd when directed at me will be correctly read as a racist attack when directed at somebody else.
And that's not even counting when they'll just come up with new terms so they can be awful in ways that are novel enough that automated filters or out-of-date moderators won't catch. E.g.: https://www.vice.com/en/article/bv88a5/white-supremacists-ha...
In short, because there's a great deal of identity-based hate in the world, identity-blind moderation ends up being an aid to the identity haters out there.
