Re:Android 13 requires devices to have more than 2GB RAM and...
by Sirened on Saturday September 10, 2022 @12:12AM
(#32787744)
> Plus there was no reporting until a threshold of multiple (10+) matches were found, so one false positive wouldn’t do it
Not even that, the photos were not visible to some trusted human party until 10+ matches were found. That is, photos were not auto reported, they were only decryptable after the threshold was crossed and then they are reported after being reviewed by a human
Google used AI to try to find CSAM, thus it flagged the father’s picture taken to send to a doctor.
Apple’s solution was to check against a database of known abuse images. It would not be capable of flagging a new photo someone took.
Of the two solutions Google’s may find more offenders, but it would clearly find FAR more false positives.
The only way for a false positive in the Apple system is for two unrelated images to have identical perceptual hashes (with multiple algorithms?). Plus there was no reporting until a threshold of multiple (10+) matches were found, so one false positive wouldn’t do it.