Security
This page is a description of Greasemonkey's security model. For tips you can apply directly to your scripts, see Security tips.
Overview
Historically, Greasemonkey would inject a user script into a page by creating a <script> tag with the user script contents inline, and appending it to the content page's DOM.
Mark Pilgrim originally described a security flaw with this design, on July 19th 2005, while Greasemonkey was at version 0.3.4. Essentially, the issue was that Greasemonkey scripts are given special permissions that the rest of the javascript running on the web page is not. For example, Greasemonkey scripts contained their own GM_xmlhttprequest object which, unlike a normal xmlttprequest object, could access any local files one one's computer or make arbitrary requests to arbitrary sites without regard for the same origin policy that typically applies to xmlhttprequest.
Unfortunately, because Greasemonkey scripts were injected directly into the page using a script tag, these objects with special permissions could be called by a script sent by the website. In other words, if you ran a Greasemonkey script on a site, the site's own javascript could access all the files on your computer!
Greasemonkey version 0.3.5 was immediately released, with all GM_* functions disabled, to plug the security hole.
To fix the security flaw, XPCNativeWrappers, a new feature of the then-in-development Firefox 1.5, were used to isolate privileged user script code from insecure content pages. Certain other changes were made, including restrictions on the GM_xmlhttpRequest method, to disallow access to local files.
unsafeWindow
Wrapping the user script environment this way creates a sandbox.
This sandbox introduces many side effects and limitations.
To allow maximum flexibility for user script authors, the unsafeWindow property was added in to the sandbox.
The window object functions as the global scope in JavaScript.
For user scripts, this global window option is in fact a "deep wrapper" of the content window.
The content window can be accessed by user scripts, but only indirectly through the wrapper.
The unsafeWindow property is a direct line to the actual content window.
Use of the unsafeWindow property should be avoided whenever possible.
Its use has the potential to open up all the original security holes that introducing the XPCNativeWrappers fixed.
When a user script relies on the unsafeWindow property, it should be included only on trusted pages, and even then is not guaranteed to be safe.
