Hacker GitHub

← All repos

AsuharietYgvar/AppleNeuralHash2ONNX

View on GitHub ↗ ★ 1,532 Python

Convert Apple NeuralHash model for CSAM Detection to ONNX.

30 mentions on Hacker News by 18 people: 3 stories, 27 comments · first 2021-08-18 · latest 2024-11-07

Mentions per year 30 mentions · busiest year 2021 (27)
20082026
  1. …out against it. Not long after the announcement, people were sharing "collisions" on Github ( https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... ) showing that fake positives would be a problem (any alarm bells?)... which then…
    comment by luuurker · 2024-11-07 · in Private Cloud Compute Security Guide
  2. …ones reported by the police using a "neural hash", which has a non-zero amount of hash collisions. https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  3. Or just make your own fake/questionable hash collisions with a script Some Guy made on Github: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX
  4. …have researchers decompiling and scrubbing through the OS to see what’s in it and what it does[0]. https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX
  5. …NeuralHash was broken by researchers. This project apparently is able to create CSAM collisions: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX If NeuralHash is really rolled out, and if this python project can really create…
  6. …different picture create a collision though. Allow me to introduce you to my posts on github: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... Where I post good looking examples of standard test images altered fairly subtly…
    comment by nullc · 2021-08-30 · in A Screeching Voice of the Minority
  7. …metric, or condition your gradient descent to prefer good looking results. In my examples ( https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... ) I used a gaussian blur to highpass the error between my starting image and the…
  8. …Please stop repeating this falsehood. Here are some examples I generated (with a link to more): https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  9. …the visual content too much. You can. Here is an example I created (with links to more): https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... I'm so tired of people suggesting that you can't. Please explain to me why you posted…
  10. …to preimage attacks: I can take random unrelated photos and make them match other hashes, https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... This means people can take child porn images and modify them so they match targeted…
  11. …exactly. This algorithm doesn't even give exact matches for the same image on different hardware. https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX Note: Neural hash generated here might be a few bits off from one generated on an iOS…
  12. …tps://old.reddit.com/r/apple/comments/p930wu/i_wont_be_pos... Here's a really high quality collision: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX//issu... Here's 2 totally different images off by a single BIT:…
  13. …[1] I've made a statement that I won't be posting any more preimages for now for that reason: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  14. …to construct preimages for Apple's neural hash function, including fairly good looking ones (e.g. https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... ) > collide with another secret hash function The supposed other 'secret'…
  15. …state agent. There is no opportunity to learn, much less challenge the content of the database. [1] https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  16. https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... I posted some examples that look like totally normal images, they're no harder to produce, you just need to noise-shape the gradient descent so that the introduced noise has a…
  17. …images of nudity could be caught in it is because This is an absurd claim. I have demonstrated ( https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... ) that it is possible to alter existing images to match an arbitrary…
  18. …images here. They don't look anything like the original despite perfectly matching the NeuralHash: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  19. …us are unlikely to be targets of this kind of attack is little comfort to those that may be. [1]: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  20. The instructions are here: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX This was discussed on HN yesterday here (and a few other front-page stories): https://news.ycombinator.com/item?id=28218391
  21. …to generate adversarial examples looks like; I don't think anyone in the big thread on the topic ( https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... ) has shared attack code for NeuralHash in particular yet.
  22. …are behind the curve, attempts to crack the hashing algorithm more efficiently are still ongoing: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... > [..] not the final implementation [..] Why on earth would you invite people…
  23. Update: The latest results now have second-preimages that are other photos, not just noise. https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... They still look pretty bad, but attacks only get better!
  24. …"Who cares? The whole point is that there will be plenty of false positives.", this is pre-image: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  25. …an ideal scenario of this technology not the reality. EDIT: Proof on the front page on HN right now https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  26. …(from [1]) instead. [0]: https://user-images.githubusercontent.com/1328/129860810-f41... [1]: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue...
  27. Apparently not that hard: https://github.com/AsuharietYgvar/AppleNeuralHash2ONNX/issue... . See also here: https://gist.github.com/unrealwill/c480371c3a4bf3abb29856c29...