Recent research suggests fine-tuning AI models for gender equity can accidentally cause extreme moral inconsistencies. For example, models consistently reject harassing a woman to stop an apocalypse ...