University of Missouri researchers developed an AI method that detects hardware trojans in chips with 97% accuracy, raising reliability concerns.