I mean even the verification is problematic. At no point I want an "certified" Apple employee or another parent looking at naked pictures of my kids, for example.

But there's nothing in this proposed implementation that could ever possibly result in that, because random pics of your kids would not be in a database of known CSAM content. So your pics wouldn't match the hash values.

I'm constantly surprised how even people on HN are confused about this - read the white paper apple published. It very explicitly says that they are using a perceptual(similarity based) hash, and we(well, not me specifically, researchers) have demonstrated that it's trivial to produce a picture that isn't even remotely similar in theme but still produces the same perceptual hash.

Apple's solution to this problem is that their employee will actually verify the picture before sending it to authorities. Which again, is one of the problems people have with this system.

If that hash stuff actually works (identifies CSAM with high probability), then that puts Apple in the position of knowingly possessing CSAM. The "One Bad Apple" article a couple days ago mentions that their staff was sometimes exposed to CSAM but that it was unknowing, because the prior probability of an image they looked at being CSAM was quite low. If the probability was high (like > 50%?), it wounded like they would have been obligated to send it to NCEMC without looking at it.

https://www.hackerfactor.com/blog/index.php?/archives/929-On...

So you are saying the hash values can never result in a collision? That in fact there is literally zero chance that two different images could result in the same hash?

Because that doesn’t sound correct to me…

It is essentially true that a cryptographic hash can never collide. But what Apple claims to be using is a perceptual hash. Perceptual hashes can not just be scaled wider trivially, to reduce probability of collision, like cryptographic hashes can.

So in other words, there is a chance for collision. And when (not if) there's a false positive, eventually the system will transmit private information to some underpaid army of people who will be receiving either real child porn or pictures of someone's naked children for confirmation.

Ignoring the nasty aspects of doing that kind of work, I don't see any way to buttress the fact that Apple has taken another step on the universally one way street of ever increasing surveillance. And whataboutism in the form of "oh but other cloud providers are already doing this" is an extremely weak argument because I don't want to use other cloud providers. I liked Apple, because irrespective of their real motives (ie money), they seemed to be privacy focused. The backhanded way they tried to sell this "feature" shows that they are just as bad as the others.