“The system could only match "exact fingerprints" of specific known child sexual abuse images, he said.”
Or like whatever they, the US govt, or any govt where they want to make money (such as China) wants. Is anyone auditing the blacklist? Is it publicly reviewable? (Since it contains CP of course not)
China can start finding Uyghurs based on the images they tend to share. If we're unlucky (as a world), they can even start searching for particular individuals.
"Save the children" is just the classic political ploy to get a ruling through that's just a precursor for evil things to come.
I'm absolutely disgusted by Apple.
They already were storing the photos unencrypted (or with keys available, at least) on their servers, so any government that was able to push them to add a hash to this scanning system could have gotten them to scan for something in iCloud.
China, in particular, could definitely already be doing that, since China made Apple host all iCloud data for Chinese users on servers inside China that're operated by a Chinese company. See: https://support.apple.com/en-us/HT208351
If Apple's willing to change the new system to do a full scan of all photos on-device and send notifications to them outside of the upload-to-iCloud-with-security-tickets mechanism, they could just as easily have done that with the old system.
What do you mean by capability here?
I fully understand the difference, but I’m just saying it’s not like Apple had to build a massively different system to support this.
Can a government secure Apple's cooperation in that? I have no idea. But it does make a useful subversion of the hash database a more complicated thing to accomplish.
Speaking for myself, if Apple is correct about those odds, I'm not personally feeling creeped out by it. If they're wrong, my opinion could change. I certainly don't have the math and security background to actually verify their claims from the white paper they posted about the system, though.
Getting more information about the hashing function they're using would be nice. It'd make it much easier to see how actually collision-prone this is. I'd be all for them getting some external review of it published, much like the review of the security-tokens they've published. (I appreciate that it's difficult, because providing the hashing function itself to experiment with lets awful people tune their images to be just-distinct-enough that they won't match.)
It's worth bearing in mind that Apple has a fairly strong motivation for the hashing to be good. They have to pay reviewers to look over these matches, and it's bad PR if it turns out that they're massively backlogged.
Where did you see that? I tried to find more info about it, but I didn't find anything.
1) That is an MD5 hash, not a perceptual hash. Apple is not using md5s.
2) It is a false positive, not bad info in the database. All involved acknowledge the possibility of false positives.
NCMEC generated the hashes Apple will use by running NCMEC's collection of forbidden media through Apple's algorithm. And perceptual hashes have more collisions than cryptographic hashes.
Several people said the database includes non CSAM seized in investigations.[1]
And if memory serves well it doesn’t need to be Apple as a whole cooperating, a single employee with oversee power could be enough.
Their approval/report rate will make a FISA court judge blush.