This is a general problem with banning things instead of simply regulating them somehow. How would one scientifically study methamphetamine, for example, in a country where it is illegal to even possess it?
Without arguing for or against CP, what if I wanted to look up evidence that use of CP leads to increase or decrease of actual child abuse, without setting off red flags everywhere? As a Psych major, that sort of thing would interest me, for example.
It's actually quite well designed and questions such as these were definitely taken into account during the design phase.
I came up with a similar scheme about 15 years ago (as a result of operating that file sharing service) and proposed to the local LE that we set up a service where a 'fingerprint' of an image could be tested against known bad images, and if an image tested positive it would be flagged for review (and on an exact match it would be automatically banned).
The local law enforcement officer thought it was a great idea but it would never fly because even the hashes of the images were considered off-limits for sharing with others and they'd have to share their database of hashes with me if I were to set this up (free of charge).
Eventually MS came up with PhotoDNA and they're too big to ignore.
Agreed, but given the fact that this feeds into the legal system in a fairly direct way and that I'm only a 'one man shop' by their definition it makes good sense to insist on doing business with a larger party.
Where it goes haywire is that they then have to trust all the employees of that larger party as well but that's logic rather than CYA.
MSFT provides a set of sample images[1], some of which return as if they matched. The sample images are not child pornography, generally pictures of celebrities (which is really strange.)
I know. I helped them beta test that here in nl. Unfortunately it's not just still images. It's also videos, and encrypted still images and encrypted videos. Photodna comes into its own once you actually have an image.
Microsoft provides an API which identifies child expoilting images.