Apple said that they are intersecting multiples databases in different jurisdiction to avoid rogue hashes and that this will be available to audit.

They also said that because it's on device, security researchers will be able to check any change to the program. (probably via the Apple Security Research Device Program ?)

[1] https://developer.apple.com/programs/security-research-devic...

How can you figure out what a neural net is trained to find? Are they releasing the data set to verify? That would be the bad images we are told it is scanning for and that would be bad.. is there some 3rd party that can do the audit?

Also now that this is a thing how effective will it be at all? Or these sick people that dumb? After all this news? I do hope they are that dumb but who knows.

> How can you figure out what a neural net is trained to find

See, it's comments like this that clearly illustrate that there is confusion, and many people are still outraged over things they don't understand. A neural net is not scanning your phone for CP. You are conflating two things. Just watch the video that you're commenting on before commenting on it.

The neural net isn't exactly searching for CSAM itself. Its role is to extract perceptual features from the image, and it is applied to both the CSAM images and your iCloud images. If those were the same to start with, then the extracted features will be the same.

As for exactly how they'll do the auditing, I'm confused as well.

Thanks. That’s a helpful bit of info I wasn’t aware of.

FYI, they also published this today: https://www.apple.com/child-safety/pdf/Security_Threat_Model...

With new bits of infos too.

That paper is being discussed at https://news.ycombinator.com/item?id=28173134.