Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The Problem with Perceptual Hashes

rentafounder.com

Post reply on HN
Print view View original

31–40 of 440 posts

Re: The Problem with Perceptual Hashes

#31

Post by vineyardmike »

bryanrasmussen wrote: post #11
celeritascelery wrote: post #8

Earlier quoted context omitted.

Test it… how exactly? This is detecting illegal material that they can’t use to test against.

Not knowing anything about it but I suppose various governmental agencies maintain corpora of nasty stuff and that you can say to them - hey we want to roll out anti-nasty stuff functionality in our service therefore we need access to corpora to test at which point there is probably a pretty involved process that requires governmental access also to make sure things work and are not misused otherwise - how does anyon…

Test it against innocent data sets, then in prod swap it for the opaque gov db of nasty stuff and hope the gov was honest about what is in it :)

They don't need to train a model to detect the actual data set. They need to train a model to follow a pre-defined algo

Re: The Problem with Perceptual Hashes

#32

Post by version_five »

judge2020 wrote: post #28
marcinzm wrote: post #23

> an Apple employee will then look at your (flagged) pictures. Always fun when unknown strangers get to look at your potentially sensitive photos with probably no notice given to you.

They already do this for photodna-matched iCloud Photos (and Google Photos, Flickr, Imgur, etc), perceptual hashes do not change that.

I'm not familiar with iPhone picture storage. Are the pictures automatically sync'ed with cloud storage? I would assume (even if I don't like it) that cloud providers may be scanning my data. But I would not expect anyone to be able to see or scan what is stored on my phone.

Incidentally, I work in computer vision and handle proprietary images. I would be violating client agreements if I let anyone else have access to them. This is a concern I've had in the past e.g. with Office365 (the gold standard in disregarding privacy) that defaults to sending pictures in word documents to Microsoft servers for captioning, etc. I use a Mac now for work, but if somehow this snooping applies to computers as well I can't keep doing so while respecting the privacy of my clients.

I echo the comment on another post, Apple is an entertainment company, I don't know why we all started using their products for business applications.

Re: The Problem with Perceptual Hashes

#33

Post by toxik »

marcinzm wrote: post #27

Given all the zero day exploits on iOS I wonder if it's now going to be viable to hack someone's phone and upload child porn to their account. Apple with happily flag the photos and then, likely, get those people arrested. Now they have to, in practice, prove they were hacked which might be impossible. Will either ruin their reputation or put them in jail for a long time. Given past witch hunts it could be decades be…

This is really a difficult problem to solve I think. However, I think most people who are prosecuted for CP distribution are hoarding it by the terabyte. It’s hard to claim that you were unaware of that. A couple of gigabytes though? Plausible. And that’s what this CSAM scanner thing is going to find on phones.

Re: The Problem with Perceptual Hashes

#34

Post by emodendroket »

rustybolt wrote: post #16

> an Apple employee will then look at your (flagged) pictures. This means that there will be people paid to look at child pornography and probably a lot of private nude pictures as well.

And what do you think the content moderation teams employed by Facebook, YouTube, et al. do all day?

Re: The Problem with Perceptual Hashes

#35

Post by zimpenfish »

stickfigure wrote: post #10

I've also implemented perceptual hashing algorithms for use in the real world. Article is correct, there really is no way to eliminate false positives while still catching minor changes (say, resizing, cropping, or watermarking). I'm sure I'm not the only person with naked pictures of my wife. Do you really want a false positive to result in your intimate moments getting shared around some outsourced boiler room for…

> Do you really want a false positive to result in your intimate moments getting shared around some outsourced boiler room for laughs?

You'd have to have several positive matches against the specific hashes of CSAM from NCMEC before they'd be flagged up for human review, right? Which presumably lowers the threshold of accidental false positives quite a bit?

Re: The Problem with Perceptual Hashes

#36

Post by Spivak »

rustybolt wrote: post #16

> an Apple employee will then look at your (flagged) pictures. This means that there will be people paid to look at child pornography and probably a lot of private nude pictures as well.

Yep! I guess this announcement is when everyone is collectively finding out how this has, apparently quietly, worked for years.

It’s a “killing floor” type job where you’re limited in how long you’re allowed to do it in a lifetime.

Re: The Problem with Perceptual Hashes

#37

Post by drzoltar »

The other issue with these hashes is non-robustness to adversarial attacks. Simply rotating the image by a few degrees, or slightly translating/shearing it will move the hash well outside the threshold. The only way to combat this would be to use a face bounding box algorithm to somehow manually realign the image.

Re: The Problem with Perceptual Hashes

#38

Post by nine_k »

stickfigure wrote: post #10

I've also implemented perceptual hashing algorithms for use in the real world. Article is correct, there really is no way to eliminate false positives while still catching minor changes (say, resizing, cropping, or watermarking). I'm sure I'm not the only person with naked pictures of my wife. Do you really want a false positive to result in your intimate moments getting shared around some outsourced boiler room for…

Buy a subcompact camera. Never upload such photos to any cloud. Use your local NAS / external disk / your Linux laptop's encrypted hard drive.

Unless you prefer to live dangerously, of course.

Re: The Problem with Perceptual Hashes

#39

Post by zimpenfish »

celeritascelery wrote: post #8
ttul wrote: post #7

Apple would not be so naive as to roll out a solution to child abuse images that has a high false positive rate. They do test things prior to release…

Test it… how exactly? This is detecting illegal material that they can’t use to test against.

> This is detecting illegal material that they can’t use to test against.

But they can because they're matching the hashes to the ones provided by NCMEC, not directly against CSAM itself (which presumably stays under some kind of lock and key at NCMEC.)

Same as you can test whether you get false positives against a bunch of MD5 hashes that Fred provides without knowing the contents of his documents.

Re: The Problem with Perceptual Hashes

#40

Post by zimpenfish »

ben_w wrote: post #9
celeritascelery wrote: post #8

Earlier quoted context omitted.

Test it… how exactly? This is detecting illegal material that they can’t use to test against.

While I don’t have any inside knowledge at all, I would expect a company as big as Apple to be able to ask law enforcement to run Apple’s algorithm on data sets Apple themselves don’t have access to and report the result. No idea if they did (or will), but I do expect it’s possible.

> ask law enforcement to run Apple’s algorithm on data sets Apple themselves don’t have access to

Sounds like that's what they did since they say they're matching against hashes provided by NCMEC generated from their 200k CSAM corpus.

[edit: Ah, in the PDF someone else linked, "First, Apple receives the NeuralHashes corresponding to known CSAM from the above child-safety organizations."]

Post reply on HN