Hacker News Reader

Comment by detaro | Hacker News Reader

The work you link is based on the NeuralHash code shipping with iOS and macOS, and explicitly says so, so there is no reason to theorize if it might already have been in a shipped OS - if it hadn't, we wouldn't have research about its abilities. (What as far as is known - and I suspect it would have been found by people scrutinizing the updates - has not been rolled out is code that actually uses it to implement the reporting mechanisms)

View on HN

Why would they distribute the ML weights file if it isn't being used for logging or reporting? Of course I do theorize that it is actually being used.

There is so much network traffic between macOS services and the internet, I'm not sure how easy it would be for researchers to discover if CSAM reporting is enabled. Little snitch detects 292 macOS services on my M1, all talking to the internet. Network experts in Apple probably even have ways of bypassing little snitch.

It could potentially be "feature flagged". The client could ship a new version with all the resources/code they need for a specific feature, but a setting on the back-end determines if the feature is enabled or not. That way, Apple could easily toggle this feature on at any point by changing the flag's value on the back-end, without needing to ship an entirely new 15.x.x binary

The weights are part of the hash implementation, so not shipping them with the rest of the hashing code would make even less sense.

Reply on news.ycombinator.com