You ever notice how the same trick that cracks a password can also fool a neural net? I was poking around the border between cryptography and ML last night.