> > intercepts the victim's notifications

> And who controls these notifications and forces application developers to use a specific service?

Am I alone in being alarmed by this? Are they admitting that their app sandboxing is so weak that a malicious app can exfil data from other unaffiliated apps? And they must instead rely on centralized control to disable those apps after the crime? So.. what’s the point of the sandboxing - if this is just desktop level lack of isolation?

Glossing over this ”detail” is not confidence inspiring. Either it’s a social engineering attack, in which case an app should have no meaningful advantage over traditional comms like web/email/social media impersonation. Or, it’s an issue of exploits not being patched properly, in which case it’s Google and/or vendor responsibility to push fixes quickly before mass malware distribution.

The only legit point for Google, to me, is apps that require very sensitive privileges, like packet inspection or OS control. You could make an argument that some special apps probably could benefit from verification or special approvals. But every random app?

> Are they admitting that their app sandboxing is so weak that a malicious app can exfil data from other unaffiliated apps?

An app can read the content of notifications if the appropriate permissions are granted, which includes 2FA codes sent by SMS or email. That those are bad ways to provide 2FA codes is its own issue.

I want that permission to exist. I use KDE Connect to display notifications on my laptop, for example. Despite the name, it's not just for KDE or Linux - there are Windows and Mac versions too.

yes, they're admitting that their APIs are powerful enough to build accessibility tools (which often must read notifications) and many other useful things (e.g. Pushbullet) that are not possible on iOS.

powerful stuff has room for abuse. I didn't really think there's much of a way to make that not the case. it's especially true for anything that you grant accessibility-level access to, and "you cannot build accessibility tools" is a terrible trade-off.

(personally I think there's some room for options with taint analysis and allowing "can read notifications = no internet" style rules, but anything capable enough will also be complex enough to be a problem)