> cannot be reverse-engineered by anyone outside of CBP to reconstruct the photo
Is this really true with GANs these days? Not that this was meant to be a technical statement of course... but I'd imagine you could train a network to create a photo that activates a given template?