IBM's AI watermarking method protects models from theft and sabotage

... against adversarial attacks that might, for example, fool a computer vision algorithm into seeing cats as "crazy quilts," or force an autonomous car to ...