Look at it from a machine learning perspective: moral hazard is the most poorly labeled dataset of all. Adding as much synthetic data to the mix as possible is the best you can do. We all acknowledge it's insufficient. But reading authoritative texts, writing as best you can, and most importantly, practicing are the only things you can do. What exactly do you want? No war? That's a political issue. No atrocities? As grandparent said, to err is human. Let he without sin cast the first stone.