About 2/3rds of their original letter was spent characterizing the system to warn kids about dick pics with on-device inference as an iMessage backdoor, which literally nobody serious believes.
More astounding was the privacy issue they raised with it, which I've not seen raised by anyone else.
Let's review how this feature works. It is only on if parent's explicitly enable it for their child's phone when they set up the child's phone for parental controls.
If it is on, images sent to the child's phone are scanned using a ML system to recognize sex images. When such an image is found, the child is given a screen that warns that the image contains content that may be harmful to the child, and may be an image of someone who did not consent to having it sent.
The child is asked if they want to reject the image or view it.
If the child elects to view it and is 13-17 they are shown a blurred version of the image and that is the end of it.
If the child elects to view it and is under 13, they get another screen that says that if they view it their parents will be notified because the parents want to be able to check to make sure the child is safe. They are again asked if they want to reject it or view it.
If they reject it, that's the end of it. If they view it they get a blurred version and the parents are notified.
The privacy issue the EFF has with this? If I send your 12 or under child a dick pic and they elect to view it knowing that their parents will be notified and see a copy of the image, my privacy might be violated because I did not consent to the child's parents being told I'm sending their child dick pics or to the parents seeing my dick pic.
I wonder what the EFF's opinion would be if I sent a dick pic to a 12 year old whose device does not have parental controls, but the kid decided to show it to the parents. Has the child violated my privacy? If we are in a state that has a civil law against nonconsensual image sharing would the EFF help me sue the child?
I have nothing to add except that once a child accepts the risk at each prompt, I believe they get to see the original image and not a blurred version.
When I first read their objection, I thought that the system would transmit the image from the child’s device to the parent’s device. I could see how that could be problematic. Except it doesn’t: the record of the image stays with the child’s device, and the parent is simply told the record has been created. At this point, the most charitable interpretation I could give is that they’re worried the model will have many false positives and ping parents about every photo a child receives. iMessage back door, this is not.
Their “concern” is literally as absurd as you describe.
