A lot of stuff falls into grey areas both for AI classification and human opinion. Let's not forget that humans disagree widely over what's pornographic and what's fair use. Let's not pretend there's a canonical oracle of pornography or (apart from court rulings) what's fair use. It stinks when their AI is clearly wrong or when you disagree with their AI, but it's hard to fault them for setting some accuracy goal when they're trying to hit a fuzzy target anyway.

Maybe your complaint is really that their appeals process is too opaque or difficult. There I agree, but I don't fault them for stopping short of perfect on their AI classifiers.