Well, obviously it could not detect *new* abuse using a hash of prior known old images, so instead it would have to extrapolate from prior images. i.e. they'll have to extend it with AI, to model a *class* of images to be able to meet their stated aim here, otherwise it could never work.
And how would you process video? Presumably most media people take these days is video. So its gonna have to be extended to cover video too. Fuzzier.
Flagged media would have to be reviewed by people. For review, you'd have to send the files off for remote review to a human, to check, (because its a fuzzy imperfect algo). So your data is no longer yours, hi-viz stalker gets remote access to the files.
Can you imagine your devices scanning your media, scanning your kids devices, sending them off to a bloke in a darkened office, who likes his job reviewing peoples private media. A stalker in a hi-viz jacket, free to browse private data, tap tap tap, reading friends messages, tap tap tap, review location data, no legal protections from said stalkers, and nobody free from their surveillance?
And no doubt the laws will get the usual "security" exception, as they all do and will lose its "for the children" veil.
Would the above happen? Well of course it would. You've established the surveillance principle, and undermined the privacy right, that was the only thing protecting people from a surveillance state. They had a stated aim, which required they extend the surveillance in order for it to work. It seems to be the whole point of this effort. And show me a privacy right that doesn't have the "security" exception to undermine it.
Biting the hand that feeds IT
