As far as I can see the claim made here is not correct. Assuming the images do as claimed match the hash, they must also match the ‘visual derivative’ in order to trigger a match. The system isn’t as easily fooled as is being claimed here.
You have misunderstood. NeuralHash is the visual derivative. Read [1] carefully, it's a very confusing document even for experts - nowhere is there a second step to this process where some second type of "visual derivative" is matched. The NeuralHash is what matters, solely. [1] https://www.apple.com/child-safety/pdf/CSAM_Detection_Techni...
Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
101–108 of 108 posts
Re: Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
#102Earlier quoted context omitted.
You have misunderstood. NeuralHash is the visual derivative. Read [1] carefully, it's a very confusing document even for experts - nowhere is there a second step to this process where some second type of "visual derivative" is matched. The NeuralHash is what matters, solely. [1] https://www.apple.com/child-safety/pdf/CSAM_Detection_Techni...
There’s literally a page showing “NeuralHash + visual derivative” in the thing you posted.
This thread was good https://twitter.com/fayfiftynine/status/1427900272148246530
Be specific, because I cannot find it.
Re: Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
#103Re: Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
#104These harmless generated images have a neuralhash equivalent to those provided in the NCMEC database submitted for testing. I repeat: Dont upload these harmless images to iCloud as Apple will assume its Child Porn (CSAM). Scripts were available on a GitHub repo but were removed because they may cause damage to others.
If Google Drive scans with the same database then how is your link working?
Re: Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
#105Re: Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
#106Earlier quoted context omitted.
Genuine question: If those image were really generated from illegal porn, are those images themselves considered illegal? Or in other words: How much do you have to modify illegal images for them to become legal again? Or do they stay illegal no matter how much you transform them?
Looking at the script below, it looks like it uses a gradient function for loss so that it learns to approach an image that generates a collision. If the case that the hashes themselves, being a result of a neural network, can be reverse engineered into pornographic images then does that raise a legal quandary? Apple said that the risk of collision is "1 in one trillion" which for a hash function would be terrible. W…
> As the system is initially deployed, we do not assume the 3 in 100M image-level false positive rate we mea- sured in our empirical assessment
The "1 in 1 trillion" part is the probability that the number of false positives could exceed the threshold needed to trigger a human review:
> Apple always chooses the match threshold such that the possibility of any given account being flagged incorrectly is lower than one in one trillion, under a very conservative assumption of the NeuralHash false positive rate in the field.
source: https://www.apple.com/child-safety/pdf/Security_Threat_Model..., page 10
Re: Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
#107EDIT: Oh, I mixed up tabs. This is a link to a google drive of pictures. Because I have scripts disabled, I got no thumbnails, and I'm thinking since this was flagged, maybe I really don't want to get any thumbnails.
Re: Tell HN: Dont upload these images to iCloud as Apple will assume its Child Porn
#108The images shown do appear to be adversarially generated inputs against some NN-based image hash or classifier, but there is no evidence to suggest that this is at all related to Apple's NeuralHash, or that the colliding hashes are from a real CSAM database (the target hashes are not public).
OP claimed they would "release 5 pieces of proof in the next 5 days" [1], and guess what, 11 days later they still haven't.
Look at OP's post and comment history, it's quite clear that they are a troll.
In the mean time, it has been actually proven that hash collisions against NeuralHash are trivially possible, see [2]