The paper you linked doesn’t reveal blurred versions of the images being hashed. It does train a classifier to determine which of 1,000 ImageNet classes an image belongs to, which “achieved a top-1 test accuracy of 4.34%”.

> to reveal blurred version of the images being hashed

Skimmed your linked paper. It seems they were able to classify hashes up to ~8% top-1 accuracy and ~30% top-10. Not exactly a blurred version, or any images at all.

So for example, they can say that you probably have images of trees, or images of buildings, but without much other data & very low accuracy.

I'd still be a lot more concerned about them simply flagging political images rather than trying to get a broad understanding of what type of photos I have

I'm pretty sure this article if from after Apple introduced (floated the idea of?) client-side scanning. I'm not sure it really bolsters the idea that it's been open for years.

This is not client side. Images just sitting on your Android phone are probably safe (although Google could push an update at any time). Your images get scanned when you back them up, send them over RCS, etc. I have even seen criminal cases originating from reverse image search - anything that touche the servers of the big tech companies, except apparently Apple, will be scanned using questionable AI and against a secret list to Protect the Children.

Google is not Apple. Apple customers expect the higher standard.

There were two different technologies. One was client-side scanning for known CSAM, which created a huge backlash and is described here: https://educatedguesswork.org/posts/apple-csam-intro/

The other is detection of images that may contain nudity, whether sent or received, when the owner/admin/parent enables the feature. It is relatively uncontroversial and is described here: https://support.apple.com/en-us/105069

That is what they actually deployed. They were planning on performing client-side scanning of all images uploaded to iCloud for CSAM and reporting it to the authorities, but backpedaled after public push-back.

The original proposal was to use a visual hash designed to identify known bad CSAM images even if cropped or otherwise edited. Your computer would scan all your stuff for these images and report you to apple who would report you to the cops. This presented a number of issues.

Accidental false positives could lead to horrific outcomes up to and including oh look bob got shot by the cops for resisting.

It was possible to produce apparently matching innocuous images and then poison people's machines with them.Oops did you click on that picture of a tree have fun with the cops. Like an advanced form of swatting.

Although inspired by a desire to find CSAM Apple could be forced to scan for ANYTHING by repressive regimes including America and China.

Although initially targeting images client side scanning of messages is a pretty obvious next step. Again obvious good motivation exists and is completely justifiable who doesn't want to stop the next mass shooting or terrorist attack... and then we can basically use it to find people critical of the regime. Do remember we are presently prosecuting a political figure for a picture of sea shells and a guy in texas is rotting in prison for distributing political literature.

Yes, in the sense that you have a legal doctor-patient privilege that binds what they can share with whom. There's not really an Apple cloud user privilege.

No, in the sense that your therapist is still required to report you to the police in various situations where you pose an immediate threat to yourself or others, etc.

Imagine you have pictures of someones baptism and the kid was nude. Is it CSAM? I think the program would have to say use but morally I'd say no. The issue with client scanning is it has to assume the worst, or they are than liable. If its the person has 20+ different baptism of nude babys well huh that actually might be CSAM because the context of how that concentrated photos implies but even than its hard what if that person actually has 20 God Children its less crazy than one thinks... Especially if they have multiply phones from a single baptism.

You might not like pictures that way but honestly I think more important in procescuting CSAM is to go after the large sources of CSAM generation. Its trafficing in East Asia, and in Europe. I think weirdly America actually produces less CSAM in general because Americans are lot more off put by Sex than most other cultures. Abuse definitely happens in the US but making policy decisions like this produces bad policy.

Does iCloud rehost the photos to other people I don't really know because I use andriod tbh. If they are being rehosted (I assume to members of your contacts) that can be problematic but I think honestly the issue a lot more complex than just protect the children which the source of critic is a lot attacks against apples are coming from

The practical problem is, without a threshold, they'd have an order of magnitude more false positives than 'real' detections, making the system useless.