At that point there is no such thing as identifying faked videos. Instead you'd rely on the abundance of independent videos and see how consistent they are. Generating a lot of them doesn't do you any good unless you can insert them into the black boxes of the 3 planes and 40 cars in the area, the databases of all the drone operators, the internal storage of all the nearby smartphones, etc.

The great part is that using a GAN means we cannot automatically distinguish a fake from a real... because that's the whole point of the GAN.