We also don't know what "human review" constitutes. It may well constitute a human simply reading the incorrect transcription and agreeing that it violates policy, without any reference to the audio track. The details of how these tech companies are overdependent on AI are not publicly known, we merely observe the consequences that can only be explained with reference to it. So I can't tell you if there was never even a human involved in the appeal, or a bad human, or one who didn't care, or whatever. We can only infer the possibility space from the outcome, not nail down the exact details.