Such a collision would generate a human validation.
This applies to cloud-shared files. It actually has applied to cloud shared photos for years. It applies to literally every major cloud photo service.
2. is it really reviewed by a human? I see how YT works, and automated failure at scale is the name of the game
3. apple has said that the on-device scanning only provides a binary yes/no on CP detection. How do you defend against a "yes" accusation? (when stored on someone else's server, the evidence is there)
There would also presumably be human review involved in the legal process, e.g. law enforcement getting a subpoena based on Apple notifying them, and then using gathered evidence for a warrant.
The system is based on known image hashes, not arbitrary ML detection.
As this system is used only for iCloud photo uploads, the evidence gathering should be similar to that done by LE with other cloud hosting providers for years
lol picky I know
We have no idea about the hash collisions. When talking about whole world, 2^256 isn't a big enough space.... even if they're using 256 bits.
And how dare anybody criticize this - criticism is tantamount to being for child porn. (Then again, that's why it was chosen. We'll soon see other things 'forbidden'.)