Simple machine learning classifiers trained on model internal states to detect specific properties like deception.