The unequal treatment of demographic groups by OpenAI content moderation system

I came across this very interesting article:

It seems to me the implementation of the moderation policy has systematic biases. I’d love to hear feedback from the developers on this seemingly big problem.