When a model learns noise in the training set and fails to generalize to new data.