> The same set of hashes is stored in the operating system of every iPhone and iPad user, so targeted attacks against only specific individuals are not possible

> The same set of hashes is stored in the operating system of every iPhone and iPad user, so targeted attacks against only specific individuals are not possible under our design. I can't figure out how they are able to guarantee this. It seems like they could take any image (e.g. one making fun of a government official) generate its NeuralHash and reverse engineer what the output of the on-device encryption would produce if that had been in the database at the time of publication. This would let them target a particular individual.