"The main purpose of the hash is to ensure that identical and visually similar images result in the same hash, and images that are different from one another result in different hashes. For example, an image that has been slightly cropped or resized should be considered identical to its original and have the same hash."
Not sure if you're being obtuse or just want to argue. In the context of Apple's implementation, it is clear they are trying to say that the same image that is cropped/resized/colour changes but still looked similar to the original image, which was supported by example by Apple. Taking the discussion out of context doesn't help in any discussion.
That's not what it says, it says similar images, and there's a lot of ways that can happen. It also says the other to make people like yourself more comfortable with the concept. It's true, a match would be very rare given the parameter that we have been told about, but those parameters can be changed quite easily. It's actually based on perceptual hashing and just allowing more differences is all neuralhash needs to match photos with only similar features.
The trick here is I don't trust Apple to do what they say now. The act of putting the scanner/hashing/matching on the phone made sure of that. And others thinking the algorithms/hash db wont change once it's there are only deluding themselves.
You're right in that its based on perceptual hashing (likely due to a cursory reading of Apple's documentation). What you're clearly confused about is that you don't understand what perceptual hashing is. It doesn't "say the other to make people... more comfortable with the concept." They are explaining, in laymen's terms, the general idea behind the concept.
Delusion is a funny thing; it's often we see others as deluded when we ourselves have misappropriated facts.
I understand exactly what it is. This part of the conversation is over, personal insults are out of bounds.
You're not alone: a lot of people here seem to understand exactly what it is without any qualifications whatsoever. It would be nice if any of the self-proclaimed experts here would at least point to some evidence for their theories.
You're welcome to disengage from a conversation you stirred up. All I did was provide some clear evidence that contradicted your claims 🤷♂️
Where precisely did I personally insult anyone?
Would be smart to wait for the sleuths to determine if the Code was removed when the gold master gets released, however Apple is probably aware of what would happen if they left it in there so I don’t expect it will be there…. Trust is pretty much gone now though
I won’t install iOS 15 only because I wanna wait for it to mature a little.
I don’t understand your logic. Apple has not released the code yet, and security researchers haven’t been able to pick it apart. Apple has been open and honest about what their plans were, yet they lost your trust?
not me. they lost my trust hard and square right here.
Well, at least, a previous version of the code was found in iOS 14.3 and can be seen here: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX
What's the problem with the algorithm, the fact that you can force a collision if you know the original file's NeuroHash? That's not really a problem. Each false positive that is found, goes through another system hosted solely on Apple's server that checks it against a different hash that is not known to anyone else, so even if you did "trick the system" with a "forced collision", it wouldn't make it through the second server-side hashing process, so they STILL wouldn't go to human review. Nobody will see your photos unless you have REAL CSAM.
It's not even that much of a problem. The CSAM hashes on the device are encrypted and unreadable by iOS, so somebody trying to engineer a hash collision wouldn't have access to any hashes to try to match.
This is even more problematic in my opinion. The problem I see is that images that are false positives leave my phone without my consent. I want to know where my files go and I do not want my phone to send the files somewhere in the background neither for another hash check nor for a human review. This has nothing to do with being scared.
If you use iCloud Photo, your image leaves your phone, regardless of whether there’s a CSAM hash done or not. If you don’t use iCloud Photo, your image stays in your phone and no CSAM hash is done if your phone is running iOS 15.
If I use iCloud Photo I consciously chose to send my photos to the iCloud Photo server and only to the iCloud Photo server for one specific purpose of which I am aware. I would even be ok with Apple scanning the files on their server.
Not sure if you understand how Apple is implementing the entire widget.
Doubters will insist there no way to know the voucher isn’t generated. While they are wrong about that, the important thing to note is that even if it is, and the ‘scan’ is performed, that ‘scan’ is effectively disabled because the positive/negative result can’t be learned on the phone. Only iCloud can divine the result. It’s like a blood test. The results aren’t learned in the needle, they are learned at the lab.
Security vouchers are generated for every image being uploaded to iCloud. An essential part of the system is that no process can ever know if there’s a match until the threshold is met on the server, so every uploaded image is treated the same way.
Apple are not obliged to search for CSAM, only to report it if they find it. I believe Apple’s intention is only to prevent a future E2EE product facilitating predators in their crimes (which is a very responsible position IMO) but you can’t block CSAM without finding CSAM and you can’t find CSAM and then not report it.
Apple have been too clever for their own good with this mathematical ethics stuff. We saw hints of it with their Differential Privacy. It’s complicated. People won’t get it on the first read and if you don’t get the messaging right, most people won’t give it a second read.
