Forgive me if I'm mistaken, but isn't the technology to detect fake video in fact part of the process of generating it? I was under the impression that these deep fakes used an adversarial network, thus, the better the detection of fakes, the better quality of the generated fakes.
Yes, but in order to be useful for training, the fake detection algorithm has to be reasonably performant. Performance is less of an issue if you just want to test if one video is fake.
Unless of course the attacker has vastly more computational power than the person trying to detect the fake
