6 ms·

While personally decidedly not a fan of this, Apple's proposed attempt on CASM was quite different from what is generally alleged in that article. In principle

While personally decidedly not a fan of this, Apple's proposed attempt on CASM was quite different from what is generally alleged in that article. In principle it matched a downloaded set of encrypted hashes of known offending images against any images before they were uploaded to the iCloud and this locally. A recorded offence would result in a local warning, and, if you proceeded, the offending image would have been exposed to further scrutiny and review (by this breaking the end-to-end encryption promise, but you would know and this would have implied your consent). It's not that far from Gatekeeper and other signature matching malware detectors. You may have already problems with the latter, but this is basically where we are.

vorpalhex 5y ago
You're depending on Apple not tampering with or receiving a tampered CSAM hash list. The same company that opened Chinese iCloud to the ccp, took down a voting app when asked by Russia and generally blows around like a leaf on the ask of authoritarian governments. Gatekeeper and app notarization went from something you could bypass to now a hard block on distributing your own software on MacOS without approval from Apple.
- smoldesu 5y ago
  The centralized root of trust is what really turns me away from Apple these days. Why should I put my faith in the world's largest corporation? Are all my eggs safer in this basket? Is this one 'too big to fail' now?
- robertoandred 5y ago
  Why wouldn't the reviewer note that the flagged image isn't CSAM?
  - mulmen 5y ago
    Why would they? What is their motivation to do so?
Clubber 5y ago
The other thing they did was use AI image recognition to scan images sent via text. This wasn't a hash match and much more dangerous. I'm not sure why people get fixated on the CASM thing while ignoring this; this seems much worse to me. The first feature of the three features will use the phone’s on-device machine learning to check the content of children’s messages for photos that look as if they may be sexually explicit. That analysis will be done entirely on the phone, Apple said, and it will not be able to see the those messages. If a child receives such a message, it will be blurred and the child wil be warned that it could be sensitive and given information about such messages, as well as being given the option to block the contact. If a child decides to view a message, they will be told that their parents will be alerted, and an adult will then be notified. https://www.independent.co.uk/life-style/gadgets-and-tech/apple-photos-messages-child-abuse-privacy-safety-b1897773.html https://www.independent.co.uk/life-style/gadgets-and-tech/ap...
- masswerk 5y ago
  > The other thing they did was use AI image recognition to scan images sent via text. I think, this would have used the already built-in Vision framework [1], which runs locally, and only on any attempts of publishing to the iCloud. This is really the conflicting point: while it should be implemented at the publishing end, the concern for end-to-end encryption prevents any such idea, for which has to be migrated to a local service. Regarding the procedures and treatment of children as receivers of such images, this is an actual (while restricted) leak, but it's a fact that minors are not recognised as independent individuals or as deserving much privacy of their own. It may be deplorable, but that's pretty much established in society and by law. Also, it may be argued that proceeding to view such an image after a warning implies consent. [1] https://developer.apple.com/documentation/vision https://developer.apple.com/documentation/vision
  - Clubber 5y ago
    >and only on any attempts of publishing to the iCloud. The feature I'm referring to happens regardless of publishing to the cloud. It scans text messages real-time and uses AI rather than a hash of known/existing images. The other feature everyone is talking about uses CSAM hashes and scans only when publishing to the cloud, but of course that feature can be easily changed with a bit switch.

vorpalhex 5y ago

You're depending on Apple not tampering with or receiving a tampered CSAM hash list. The same company that opened Chinese iCloud to the ccp, took down a voting app when asked by Russia and generally blows around like a leaf on the ask of authoritarian governments. Gatekeeper and app notarization went from something you could bypass to now a hard block on distributing your own software on MacOS without approval from Apple.

smoldesu 5y ago
The centralized root of trust is what really turns me away from Apple these days. Why should I put my faith in the world's largest corporation? Are all my eggs safer in this basket? Is this one 'too big to fail' now?
robertoandred 5y ago
Why wouldn't the reviewer note that the flagged image isn't CSAM?
- mulmen 5y ago
  Why would they? What is their motivation to do so?

Clubber 5y ago

The other thing they did was use AI image recognition to scan images sent via text. This wasn't a hash match and much more dangerous. I'm not sure why people get fixated on the CASM thing while ignoring this; this seems much worse to me. The first feature of the three features will use the phone’s on-device machine learning to check the content of children’s messages for photos that look as if they may be sexually explicit. That analysis will be done entirely on the phone, Apple said, and it will not be able to see the those messages. If a child receives such a message, it will be blurred and the child wil be warned that it could be sensitive and given information about such messages, as well as being given the option to block the contact. If a child decides to view a message, they will be told that their parents will be alerted, and an adult will then be notified. https://www.independent.co.uk/life-style/gadgets-and-tech/apple-photos-messages-child-abuse-privacy-safety-b1897773.html https://www.independent.co.uk/life-style/gadgets-and-tech/ap...

masswerk 5y ago
> The other thing they did was use AI image recognition to scan images sent via text. I think, this would have used the already built-in Vision framework [1], which runs locally, and only on any attempts of publishing to the iCloud. This is really the conflicting point: while it should be implemented at the publishing end, the concern for end-to-end encryption prevents any such idea, for which has to be migrated to a local service. Regarding the procedures and treatment of children as receivers of such images, this is an actual (while restricted) leak, but it's a fact that minors are not recognised as independent individuals or as deserving much privacy of their own. It may be deplorable, but that's pretty much established in society and by law. Also, it may be argued that proceeding to view such an image after a warning implies consent. [1] https://developer.apple.com/documentation/vision https://developer.apple.com/documentation/vision
- Clubber 5y ago
  >and only on any attempts of publishing to the iCloud. The feature I'm referring to happens regardless of publishing to the cloud. It scans text messages real-time and uses AI rather than a hash of known/existing images. The other feature everyone is talking about uses CSAM hashes and scans only when publishing to the cloud, but of course that feature can be easily changed with a bit switch.