That is extremely troubling. I wonder if it's because the algorithm is taking into account previous negative uses of the word in comments/commentary and flagging it as possible hate speech?
This is the point that various people such as Maciej Cegłowski have been making - if they had a policy which banned that word, we could argue about it; but if they have an algorithm that emits the same result but without explanation, it's much harder to make criticism stick. And unless people go to special effort, the algorithm will pick up the prejudices of the training data.