a lot of this stuff is already generated using generative adversairal networks, so i think the AI to fool is just as good as the one thats trying not to be fooled by design.