Thread by @matthew_d_green

I’m glad that Apple is feeling the heat and changing their policy. But this illustrates something important: in building this system, the *only limiting principle* is how much heat Apple can tolerate before it changes its policies. reuters.com/technology/aft…

This headline pretty neatly summarizes Apple’s progress this week. gizmodo.com/apple-will-kee…

I’m grateful that Apple has been so open and positive to the technical community. I wish they’d done this before they launched their unpopular service, not after. Some of us have been talking about these issues for two years.

People keep asking whether clarification of the technology will help. I think the problem here is that the technology details actually *obscure* what this system does, and Apple is using this to have it both ways whenever they get criticism.

For example, when Apple initially launched this last week they were clearly worried that their system would be mistaken for server-side scanning. So they emphasized that it scans photos *on your device.*

This created a backlash because people (IMHO rightly) asked: why do you need to scan all the photos on my device? At which point Apple began to “clarify” the issue by pointing out that there was a complicated two-party protocol and so maybe matching doesn’t happen on the client?

In writing this op-ed for the Times, @alexstamos and I had to go through two days of fact checking delays while Apple kept insisting that their system doesn’t do scanning on the client. We kept quoting back their own documentation in response. google.com/amp/s/www.nyti…

Apple has also started emphasizing that they will include “hash publications” to prevent selective targeting of individuals. This is great! None of these things were properly described in the technical/security reviews they did last week!

But overall I feel like a lot of Apple’s technical clarification has been designed to obscure what they’re doing, to people who generally understand it and just don’t like it.

In cryptography we have the notion of an “ideal functionality.” It’s a way to describe what we want from a complicated protocol by ignoring the crypto and just imagining we have a magical trusted server doing the work. Here is my impression of Apple’s last-week version.

This all assumes the crypto works perfectly, but the point here is that “is the scanning happening server or client side” isn’t the right question. The question should be: is Apple scanning photos you uploaded to the server? The answer is “no”, they’re scanning everything.

But even this description isn’t accurate, because the new threat model Apple outlined this week actually expands the ideal functionality as follows. apple.com/child-safety/p…

It now acknowledges that there are millions of devices and they all need to be scanned using the same database. (Not shown here, devices’ photo libraries can change over time too.) It also adds NCMEC, slightly breaking the formalism.