Data Leaking is when the model fits to data the model shouldn’t have access to during training.

This usually results in overly optimistic performance estimates.