Ask HN: Can deep learning be used to extract a payload from a stego image?

I've done a bit of research. Current (public) steganalysis methods merely detect steganography. Some research papers mention "locating steganalysis," which is basically finding where exactly the hidden message is in the image/stego item. After it's found, the problem becomes a cryptanalysis problem. Can deep learning be used for locating steganalysis as well as cryptanalysis? And could this be used for any stego image? Or will it be rendered obsolete as steganography software evolves?