Preferred but Less Safe? Human Preference–Safety Conflict in Paired AI Comparison Data. AI and Wisdom, v. 1, n. 3, 9 Sep.2026.