2 ms·

I am strongly against Apple’s decision to do on-device CSAM detection, but: wasn’t there a secondary hash whose database is not shared? In theory you need to co

by jchw 5y ago

I am strongly against Apple’s decision to do on-device CSAM detection, but: wasn’t there a secondary hash whose database is not shared? In theory you need to collide with both to truly defeat the design, right?

PBnFlash 5y ago
You just need a sample of something that is evidently so pervasive we're building a nation wide dragnet to stop.
- jchw 5y ago
  I doubt finding it would really be that hard, if you wanted to be on a list somewhere in some government database, but even armed with a full image that is in the NCMEC database, the problem is that the second hashing algorithm runs on the server and presumably has secure-by-obscurity details… so it would be hard to collide with it on purpose unless you are an insider. That’s my understanding, although details have been a bit shaky at times.
only_as_i_fall 5y ago
I don't see how. They're hashing on feature space (so trivial cropping and such doesn't defeat this) but they have two totally separate methods of matching those hashes? Doesn't sound right to me...
- jchw 5y ago
  Apparently the images in question would get sent to the server, and all calculation happens there. > In a call with reporters regarding the new findings, Apple said its CSAM-scanning system had been built with collisions in mind, given the known limitations of perceptual hashing algorithms. In particular, the company emphasized a secondary server-side hashing algorithm, separate from NeuralHash, the specifics of which are not public. If an image that produced a NeuralHash collision were flagged by the system, it would be checked against the secondary system and identified as an error before reaching human moderators. https://www.theverge.com/2021/8/18/22630439/apple-csam-neuralhash-collision-vulnerability-flaw-cryptography https://www.theverge.com/2021/8/18/22630439/apple-csam-neura... For one reason or another Apple really wants to create this precedent, so it’s only natural they’re doing every last thing to make the feature hard to defeat.
  - ec109685 5y ago
    Hard to exploit is better phrasing.
zepto 5y ago
Yes, but it’s much easier to just ignore that and proclaim how weak Apple’a system is.

by jchw 5y ago

PBnFlash 5y ago

You just need a sample of something that is evidently so pervasive we're building a nation wide dragnet to stop.

jchw 5y ago
I doubt finding it would really be that hard, if you wanted to be on a list somewhere in some government database, but even armed with a full image that is in the NCMEC database, the problem is that the second hashing algorithm runs on the server and presumably has secure-by-obscurity details… so it would be hard to collide with it on purpose unless you are an insider. That’s my understanding, although details have been a bit shaky at times.

only_as_i_fall 5y ago

I don't see how. They're hashing on feature space (so trivial cropping and such doesn't defeat this) but they have two totally separate methods of matching those hashes? Doesn't sound right to me...

jchw 5y ago
Apparently the images in question would get sent to the server, and all calculation happens there. > In a call with reporters regarding the new findings, Apple said its CSAM-scanning system had been built with collisions in mind, given the known limitations of perceptual hashing algorithms. In particular, the company emphasized a secondary server-side hashing algorithm, separate from NeuralHash, the specifics of which are not public. If an image that produced a NeuralHash collision were flagged by the system, it would be checked against the secondary system and identified as an error before reaching human moderators. https://www.theverge.com/2021/8/18/22630439/apple-csam-neuralhash-collision-vulnerability-flaw-cryptography https://www.theverge.com/2021/8/18/22630439/apple-csam-neura... For one reason or another Apple really wants to create this precedent, so it’s only natural they’re doing every last thing to make the feature hard to defeat.
- ec109685 5y ago
  Hard to exploit is better phrasing.

zepto 5y ago

Yes, but it’s much easier to just ignore that and proclaim how weak Apple’a system is.