This is easily defeated by pre-processing as is commonly done before training.