I’ve ranted before about how insecure web browsers are, because they trust themselves, their libraries and user-added plug-ins too much.  At a very high level, they have responsibilities that can be likened to those of operating systems, because they run potentially dangerous code from different sources (users vs web sites) and need to do it separately from each other and from root (the user account running the browser), i.e., securely.  The web browsers of today look as ridiculous to me as the thought of using Windows 95 to run enterprise servers.  Run an insecure plugin , get owned (e.g., Quicktime).  Enable JavaScript, VBScript, ActiveX, Java, get owned.  Get owned because the web browser depends on libraries that have more than 6-month-old vulnerabilities (1-year old depending on how you count), and the whole thing collapses like a house of cards.  As long as they are internally so open and naive, web browsers will keep having shameful security records and be unworthy of our trust. 

IE 7’s protected mode needs to be acknowledged as a security effort, but CanSecWest proved that it didn’t isolate Flash well enough.  It’s not clear if a configuration issue was involved, but I don’t care—most people won’t configure it right either then.  IE 7’s protected mode is a collection of good measures, such as applying least privilege and separation of privilege, and intercepting system API calls, but it is difficult to verify and explain how it all fits together, and be sure that there are no gaps.  More importantly, it relies heavily on the slippery slope of asking the user to appropriately and correctly grant higher permissions.  We know where that leads—most everything gets granted and the security is defeated.

Someone not only thought of a proper security architecture for web browsers but did it (see “Secure web browsing with the OP web browser” by Chris Grier, Shuo Tang, and Samuel T. King).  There’s a browser kernel, and everything else is well compartmentalized and isolated.  Similarly to the best operating system architectures for security, the kernel is very small (1221 lines of code), has limited functionality, and doesn’t run plug-ins inside kernel space (I’d love to have no drivers in my OS kernel as well…).  It’s not clear if it’s a minimal or “true” micro-kernel—the authors steer clear of that discussion.  Even malicious hosted ads (e.g., Yahoo! has had repeated experiences with this) are quarantined with a “provider domain policy”.  This is an interesting read, and very encouraging.  I’d love to play with it, but I can’t find a download.