I hear ya, but one could set different operating thresholds: auto-approve low-risk posts, hold ambiguous posts for review, and automatically reject very high-confidence violations. So HITL for sure, but MUCH less H in the L.