What I'm suggesting is not searching based on metadata, that sounds pretty useless, it's using image recognition and AI to have a program (rather than a human) say, I've seen that before. It would be the same kind of technology that is used for facial recognition and categorization of people's holiday photographs.
Another thing one could try, is to use ML to recognize when a piece of art for sale on Ebay is a fake copy, because someone other than the seller has it in their possession.
You could find this out by what's on CAF. Since not everyone wants their stuff up for public display, this would be done best if people could make their art available for ML without making it available for public display.
Of course, this now provides a compelling reason for someone to take their stuff off CAF once they no longer own it.
Before everyone gets their hopes up that recognition of fakes can be completely automated, it should be recognized that image recognition can be fooled if you know what you're doing, by messing with the image pixels in ways that are not detectable to the human eye. But:
The fakers that have been exposed are not very sophisticated.
Perhaps image recognition will get better (we don't really understand what it does now).
The ML stuff is maybe a bit ambitious. It would be nice to at least have a central repository of images of fakes, indexed by artist, this is not technically hard, it just requires some resources (maybe a Kickstarter?).
I don't know how widespread fakes are, but it is possible it is much more prevalent than people think, and only the amateur hour fakes are being exposed.
What really worries me are Artists Editions, all those high resolution scans of valuable art....