The Google model of an auto-updating but un-QA'd app store doesn't work for me, because it combines two things I really don't see as compatible: 1) low-friction updates; and 2) installation of arbitrary un-reviewed code from the internet. If you're going to do #2, then I want the friction of downloading a new executable. I want to go to a website, see if the company still exists, read the release notes, generally be cautious about installation of random executables off the internet. But if you're going to do #1, then since the updates are supposed to apply without significant review by me, someone else has to be vetting what goes into the repository for at least minimal non-evilness standards.
I agree on this.
The best solution for now would be a meta-extension that checks if you have compromised extensions installed and disable them.
The blacklist could be compiled based on the Store feedbacks (ratings dropping sharply? disabled.), a reporting system from the app, and also using automatic testing. For example run the extension on a sandboxed machine and check for requests to known shady domains.
I uninstalled noscript years ago when it started updating twice a week with apparently miniscule changes just to pop up its ad filled landing page. Its a fucking javascript blacklist/whitelist app, why on earth does it need to update? If I wanted a "browser security suite" (what noscript currently bills itself as) I'd download that specifically.
Sites like pinterest have extensions that request these permissions, when they don't even need them. There's a way to have the extension only have access to the site you're on when you CLICK somewhere in your toolbar.
There are fine grained permissions and optional permissions to specific hosts and ports and URL patterns. It's all very well thought out.
The problem is with users.
For example, I have an extension that enables autocomplete on all web pages: https://chrome.google.com/webstore/detail/autocomplete-on/gd..., and it has access to all data on all websites. People install it every day, I don't know why. I would never install or trust anybody else with such an extension. I have to manually clone any extension I like and upload it to the chrome web store; that way I know it's not going to auto update and do nefarious things.
With how the update system works[1], when requesting new permissions the extension is disabled until manually reenabled. If there's even a slight possibility you may want to request access to additional sites in the future, you basically have to request "all websites" to prevent this from happening.
Since Chrome 16 there have been optional permissions around (so you only request permissions when they're needed, preventing the extension from auto-disabling), although that introduces additional overhead beyond simply requesting everything in the manifest file.
The extension update system should probably work more along the lines of that in Android - it auto-updates if the new version needs no additional permissions, but requires user input when new permissions are required. The current[1] state of auto-disable-on-update isn't ideal from either a developer or user position.
[1] as of about a year ago when I last tested this
Still, I feel that using optional permissions and pointing out to the user why they would want enable the new permissions is the best option. Yes, it's more work for the developer.
I just released an extension to help identify high risk installed extensions: https://chrome.google.com/webstore/detail/privacy-guard/edca...
Of course, then you need a solid extension update system where there's some way to alert the user of when an extension needs new permissions, and have them see and authorize them before the update goes active. Hopefully something like the way that Chome updates work now - instead of throwing a modal dialog at the user at some random time, have a tick or light in the UI somewhere that they need to do something.
I'd lean towards using a finer-grained permission model for extensions, and javascript in general. As far as I know, most extensions now can only request permissions to run arbitrary JS on every page you visit, and send arbitrary AJAX to any URL. Perhaps we could do a better job of locking that down. Have no such thing as a permission to AJAX any URL, but instead make AJAXing any particular TLD a new permission. Maybe we can also only allow access to DOM elements originating from particular domains. Then you could keep auto-update, but require a new user authorization for any new URL/domain access that an extension needs. It'd probably get pretty complex, but I think we could put together something reasonable. But then the other problem is the pile of plugins out there already with those arbitrary permissions. Maybe you could let them stay, but not allow any updates without switching to the new permission model?