If an AI model can tell if an image is faked, it can tell another AI model that an image is faked. False detections are inevitable and as the line gets blurrier

If an AI model can tell if an image is faked, it can tell another AI model that an image is faked. False detections are inevitable and as the line gets blurrier as models learn to do their jobs better, false detections will become more frequent. There really isn't any outcome where it's plausible that they will become less frequent. So that isn't going to be an arms race, it's going to be a massacre. The cryptography stuff also has the potential to end badly, you can end-to-end cryptographically sign all you want but inputs are still analog, humans have been faking things "in-camera" for a long time and there's nothing stopping anyone from reading ChatGPT from their phone and typing it into Word on their laptop. So sure you have the verification that something wasn't altered digitally, but now you have to ask yourself whether you trust that every photon that hit the photosensor is authentic, that every sentence typed by a human was conceptualized by one.