"We Can't Afford to Become Idiots": How a Professor's Experiment Reveals the Scale of Academic Fraud in AI