Professors flagged AI answers as pedagogically misleading or harmful just 3.5% of the time, against 12% for peer-written ...