It's quite easy to remove "race" as a feature in the program. The more realistic and difficult question is what about features that are correlated with race, but are not race itself, like: zip code, or proportion of spending used on clothing.