This post describes a tokening system for all JSON HttpGet and HttpPost AJAX calls:
In short, you use the AntiForgeryToken attribute to create a token on the page, and then manually validate that that value is sent back to the controller via the AJAX call.
http://www.asp.net/web-api/overview/security/preventing-cross-site-request-forgery-(csrf)-attacks
After hours of internet sleuthing, there are always references to this possibility, but no one actually implements it. Instead the commonly repeated techniques are 1) using only HttpPost AJAX requests (which breaks REST), 2) wrapping all json responses in an anonymous object, which results in significantly inelegant code in .net MVC4, 3) using an unparsable cruf, which breaks common libraries such as backbone.js
So why aren't we all using the tokening system linked above? What am I missing?
If you're concerned about CSRF, and your concerned about REST and doing REST correctly, then you shouldn't be doing anything in a GET that would be affected by a CSRF, since the entire purpose of AntiForgeryTokens is to deal with changing data (ie, you can't use an AntiForgeryToken without first getting the page anyways that contains the token).
So, saying using POST "breaks REST" seems to be misunderstanding what you're using the token for in the first place. It's true that a GET can expose sensitive information, but you have to have some way to get the token first if you want to use it with GET.
The real problem with Ajax and AntiForgeryToken with json and Ajax is that the "built-in" validation only works with form values, not json. So you have to do the validation yourself, and that article you linked to gives a good explanation of how to do that.