Worse. Jane Someone sees what happens to Joe and decides to be more careful. Now Jim abuses kids more frequently to meet Jane’s need for fresh content which she perceives as safer.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Outlines Security and Privacy of CSAM Detection System in New Document
- Thread starter MacRumors
- Start date
- Sort by reaction score
You took this out of context. Follow the conversation and you’ll realize that I’m saying that CSAM-criminals will not save csam-pictures, from the internet, in their camera roll but instead hide it (criminals usually hide..)
This implementation will only catch the stupidest of the stupid criminals. These would have been caught way earlier by publicly harrasing children at Walmart etc.
This will not save the children. And at a cost of decreased privacy for the average user.
Lose-Lose.
Last edited:
No one seriously believes this won't be abused. It will be abused by tech corporations and governments alike, although they are nearly one and the same at this point. The “save the children” angle is just to get a foot in the door. This kind of thing will lead to the total eradication of privacy and, eventually, free thought.
No the government can’t engage private entities to do unconstitutional work.
See post #626
Apple kisses the ring of every government official it can. Tim Cook’s not going to risk his connections in China or other authoritarian countries that demand Apple search for more. Apple, as large as it is, cannot overcome an authoritarian nation-state, and more importantly, they’ve never bothered to try. They’ve given authoritarian governments whatever they have asked for without question, you know, “complying with local laws”
Apple is only concerned about privacy within liberal democracies’ borders, and even then, only until a law gets passed that gives them cover.
Apple can’t have those great earnings calls every quarter without the Chinese market. The moment China says you do what we say If you want to sell in our Market, I guarantee there is not a single Apple executive that would have the integrity to walk away, especially the bean counter at the helm.
Apple is only concerned about privacy within liberal democracies’ borders, and even then, only until a law gets passed that gives them cover.
Apple can’t have those great earnings calls every quarter without the Chinese market. The moment China says you do what we say If you want to sell in our Market, I guarantee there is not a single Apple executive that would have the integrity to walk away, especially the bean counter at the helm.
Last edited:
If Craig is to be believed, moving this to the device increases privacy, in that prior to this method, all of your actual photos were being accessed and scanned directly by Apple (and any government agency who got access to it). With on-device matching to cross-verified through multiple jurisdiction CSAM hashes, only those photos matching verified CSAM will be reviewed by anyone. This preserves user privacy. Also, US national security letters cannot get data, for ref: https://en.wikipedia.org/wiki/National_security_letter
In other words, they can check if you called a child pornographer, but cannot know what you spoke about or about the thousands of pedo-porn you have, unless you upload it to iCloud.
It seems many here think that routinely scanning all of your iCloud photos is somehow less invasive than the statistically improbable event in which Apple reviews a subset of derivatives of your photos that seem to match many copies of unique instances of child porn verified by multiple jurisdictions to be a true hash of real child porn (and not some nation-specific-hunt photo).
It seems this stems from the thinking that "on-device scanning" implies it is Apple itself scanning, rather than your device, which has already self-scanned for over a decade. Alternatively, it seems many think that this opens up some pandora's box/starts a slippery-slope of sorts, in which anything can be searched for if forced by local jurisdictions to do so, even though the technology, as described, would be incapable of doing this even with significant engineering efforts.
And all of this is coming from an echo chamber of many who used to think Apple had their backs in terms of privacy. The ease of the swaying of the collective mind is certainly interesting.
It seems this stems from the thinking that "on-device scanning" implies it is Apple itself scanning, rather than your device, which has already self-scanned for over a decade. Alternatively, it seems many think that this opens up some pandora's box/starts a slippery-slope of sorts, in which anything can be searched for if forced by local jurisdictions to do so, even though the technology, as described, would be incapable of doing this even with significant engineering efforts.
And all of this is coming from an echo chamber of many who used to think Apple had their backs in terms of privacy. The ease of the swaying of the collective mind is certainly interesting.
Neural hasing is based on similarities not exact matches (it’s even on Apples own white papers). For example, police might be looking for anyone who has picture of victim, suspect or any person of interest. Add location data and access all the documents (not just pictures) and you have very nice mass surveillance network. That would be Prism 2.0 right there. One could build whole case with that data.
But the system is there to detect exact matches (and derivatives).
What would an image of child safety look like?
So if they provided hashes to that to Apple they would only catch people who had the images the government had of a particular child wearing seatbelts or similar.
The system can't find images of similar nature.
unless the photos are in the database there will be no match, in other words, if you make a childporn video or series of photos and store those photos on your phone and they are uploaded to icloud they will pass through just fine
the system does nothing about NEW childporn/csam images, only matching know images stored in the csam database, in fact the images must be on at least 2 databases to qualify to get in the database
the married couple presumably would be in no databases at all
Last edited:
With Google Workspace you have better control over privacy than with Apple products (especially after iOS 15). Sure, you end up paying over $100 a year to get that privacy with Google but no one said it would be free or even cheap.
No, you haven't understood how NeuralHash works.
It's trying to find exact photos and derivates from that exact photo, not similar photos. It is optimised to not find similar photos.
Let's say you are killed by a friend. This (previous) friend has a picture of you drinking a beer in his iCloud Photo Library.
The police gets pictures of you from family members and others who knew you. Maybe they get 100 images of you. The police forces apple to NeuralHash these 100 pictures and apply them to iOS.
NeuralHash will not catch, with a high probability, your killer because those 100 pictures are not exact matches (or derivates with small changes) of the picture your killer has on his phone.
Part of NeuralHash is to use a neural network (machine learning) to also teach the system to not recognise images which are not exact or derivatives.
That's why this system is so useless in many cases to misuse.
Yes, but in a newer decision they said the search was legal because of the good-faith exception.
Also this case also involved AOL sending over additional information which they didn't know where child pornography. Apple will only send over images they know with a very high probability is illegal.
Becuase
- If on Device, only the device (not Apple) knows what it is scanning and Apple is not sent any information on the user unless the user is a pedophile.
- On Cloud, each scan is done with Apple knowing who owns the scanned image hash.
Because when doing it on client side you have infinite scalability. The client literally pays for the device which is doing the scanning and heavy lifting. Cloudflare has stated that one of the biggest hurdles of using sophisticated algorithms and AI in general to scan for child porn are the processor requirements. Also once you scan photos on device you can easily develop the system to scan other types of data also.
This is one reason. Another might be Apple doesn't want to host CSAM at all to begin with.
False --- this is in direct contrast to how the system was actually designed to work. It would take a complete removal of NeuralHash and a replacement with a more generalized neural detector, which wouldn't be necessary. They could just use the existing scanning they have been doing for over a decade for face detection, in which case why waste engineering hours to develop NeuralHash at all?
How does any layman know what's in the database? Who decides? Are Sally Mann's photos in there? If so, how many of them?
Are you just using your imagination trying to find unlikely scenarios?
A National Security Letters (NSL) may be challenged in federal court although it's not easy. Also NSLs can only be used for transactional records and other non-content, so Apple is not required to change their system to give more data to the FBI.
The people the FBI would be after would need to have on the order of 30 of these exact images (or derivates from these exact images). Can you describe what sort of images and cases this would be useful for?
Also, the system won't scale. Every time FBI served a new NSL and provided new hashes to Apple, Apple would have to create a new version of iOS and deploy that. If they did that too often, people would become suspicious.
No layman knows. If Apple is to be believed, the database contains hashes of CSAM that has been verified as actual CSAM (i.e., not propaganda, etc.) from multiple separate non-cooperating governments. It's supposed to have only actual CSAM, and the multiple jurisdiction is meant to thwart a nefarious actor from injecting material to track local citizens in some way.
You are wrong. Derivative and similar in grand scale of things is the same when source material number is high enough. In Apple white paper they are literately showing how they are matching the single colour image to grayscale image. This information is used in AI training process which consist of aprox. 200 000 images. You do understand that the whole CSAM database which Interpol is using has over 2.7 million images. You honestly believe that a) Apple is only looking for 200 000 images because those are the images that were used to train the AI. Tell me why are they only looking for such a small number of images and leaving out 2.5 million? B) Or after 200 000 child porn images the AI has enough info to recognise also the other 2.5 million images? C) This means the AI can recognise and correctly identify material outside the teaching material. D) Why wouldn’t they search child porn outside the base set? This might actually save children. With adjustments of Neural hash threshold this can be easily achieved.