Large Language Models Outperform Humans in Fraud Detection and Resistance to Motivated Investor Pressure