- one is parental control, it’s upon request, uses ML, it’s “local” (that is, it’s sent to the parent)
- the other one is plain hash matching, which does not “save the children” but rather is “catch the viewer” — exclusively. This has no impact on the abuse of the CSAM subjects because it only matches publicly-known content.
I don’t know why NCMEC is excited since stopping the viewer does not stop the abuse; This does not affect them.
Conspirationally speaking, it almost feels like things aren’t the way they’re described and Apple will in fact ML to detect new content and report that.
The threshold thing doesn’t even make sense otherwise. One known CSAM picture should be enough to trigger the report, but it sounds like they want better accuracy for ML detection.