When a machine learning model learns the training data too specifically, including noise and irrelevant patterns, resulting in poor performance on new, unseen data.