Researchers found that AI models can judge rule violations more harshly than humans. Machine-learning models, often used to make decisions about rule violations, are failing to replicate human judgment, according to a study conducted by researchers from MIT and other institutions.