I put GPT-5.2 through a 14-round test, and the AI model raised some serious questions ...