[ducadveritatem]

The "algorithm" isn't some sort of neural network trying to intelligently identify things that "look like" CP. It's a perceptual hash matching against a database of known CP. It has to find multiple matches before it flags the account for review to reduce false positives. Only after review confirming a match to known exploitative images is the info referred to NCMEC for action.

Full whitepaper: https://www.apple.com/child-safety/pdf/CSAM_Detection_Techni...

[r00fus]

Who validates said database of "known CP"? How do we know the pictures are actually not "find dissident"? Who watches the watchers?

This entire endeavor hands the keys to an unaccountable police state.