59a34eabe31910abfb06f308
Security and privacy
perceptual hash collision demo
SHOW HN2021
How HN took it
5 positive, 23 neutral or asking, 2 mixed, 32 critical, out of 62 comments about the project. Thread score -0.34.
Privacy: 4 for, 29 againstIdea: 1 for, 1 againstExecution: 0 for, 2 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Demo is a clearer explanation of the danger than long articles
From 1 comment
This simple site is a far better demo and explanation of the extreme danger of Apple's proposal than any of the long articles written about it. Thank you for caring enough to put…
smt88 on HN
Useful for informing people about the proposal
From 1 comment
Thank you for your kind words. This situation is something close to my heart, I think privacy should be a fundamental right for everyone, and it is essential to inform people.
mono-bob on HN
Objections
Human review of low-res derivatives could let false positives through
From 3 comments
The fact that they send a visual derivative of the image, such as a low-resolution version, makes the human review process tricky. Because there is, understandably, no actual…
mono-bob on HNApple recieves a visual derivative of the image, such as a low-resolution version, for manual review. This makes the human review process tricky. Because there is,…
mono-bob on HNNot even that. Apple recieves a visual derivative of the image, such as a low-resolution version, for manual review. The downside it that this makes the human review process…
mono-bob on HN
On-device scanning violates privacy and could be extended for censorship
From 3 comments
It could flag you as a supporter of Free Hong Kong in China, women's rights in Afghanistan, or maybe just someone who likes gay porn in Russia. Building these types of systems is…
echelon on HNThe point is that its done on our end, not apple's end. Many reasons this is bad. Obviously the "Slippery slope" argument around it scanning more in future and being abused by…
vineyardmike on HNSeems like this CSAM tech could be super useful in China for detecting winnie the pooh or other evidence of thought crime against the regime. Even if Apple doesn't end up rolling…
umvi on HN
Perceptual hashes are susceptible to collisions
From 2 comments
That's just incorrect. From the whitepaper: > Once Apple’s iCloud Photos servers decrypt a set of positive match vouchers for an account that exceeded the match threshold, the…
prophesi on HNEach image on the left has a blob vaguely similar to the highlights in the dog image on the right. Likely the "perceptual" algorithm isn't "perceiving" contrast the same way…
WesolyKubeczek on HN
Requests
Make the attack work on the fly for user uploads
From 1 comment
Cool! Now do one where the user uploads the image and it tweaks it to find a collision on the fly.
drexlspivey on HN
Auditable hash database
From 1 comment
Apple have stated that they will make the database of hashes that their system uses auditable by researchers. Does anyone know if that has happened yet? Is it possible to view…
spuz on HN
Questions
How are collision images found or created
From 1 comment
Can somebody please explain to me how one can go about finding images that have collision hashes? Or how you can create an image to have a specific hash?
seanbarry on HN
Who controls the hash database
From 1 comment
> For example, it's possible to detect political campaign posters or similar images on users' devices by extending the database. So who controls the database?
DavideNL on HN
Alternatives named
Google and Microsoft cloud hash matching for CSAM
From 1 comment
Now let's create one for the hash matching that Google, Microsoft, and other cloud providers use. If your problem with Apple's proposal is the fact they do hash matching (rather…
slg on HN
Apple's own iCloud scanning
From 1 comment
Bugs are kinda slippery slope arguments as well, since they are impossible to prevent or measure with 100% accuracy. These are stopped in human review. (Hopefully) It seems that…
nicce on HN