The problems occur because, unfortunately, there are an awful lot of devils hidden inside the details.
An application's "attack surface" approximates the ways in which an attacker can cause damage to your application or its users. The more technologies you use, the bigger your attack surface.
Here, then, are three simple ways of reducing your AJAX application's attack surface:
Know what runs where
Imagine you've carefully coded rules to be sure that administrative functions are never shown to ordinary users. This sounds good, but you forgot that the user interface code is running on the client. So the attacker uses Firebug to invoke the administrative functions. If the proper checks aren't in place on the server side, the attacker just gained administrative rights. Many Rich Internet Application (RIA) frameworks also have this issue.
The solution is to be very careful about making the boundary between the client and the server very clear.
Keep data separate from code
Hackers frequently use a technique called "injection" to bury commands inside innocent-looking data and get interpreters to execute their commands for them. This simple trick is at the heart of many security attacks including SQL injection, command injection, LDAP injection, XSS, and buffer overflows. Preventing injection in a target-rich environment like the modern browser takes discipline.
The key to stopping injection attacks is never executing data that might contain code. But with AJAX, lots of data and code get passed around and mashed together in the DOM. There has never been a data structure that mixes together code and data more than modern HTML.
Encoding makes everything complicated. Attackers can hide their attacks inside innocent-looking data by encoding it. Back-end systems may recognize the encoding used and execute the attack. Or they may decode the attack and pass it on to a system that's vulnerable to it.
Attackers may use multiple different encoding schemes, or even double encode to tunnel their attacks through innocent systems. There are dozens and dozens of encoding schemes and no way to tell which schemes will be recognized by the interpreters you're using. This makes recognizing attacks very difficult, if not impossible.
Every time you send or receive data both sides have to know the intended encoding. Never try to make a "best effort" attempt to guess the right encoding. You can't prevent an attacker from sending data with some other encoding through the channel, but you don't have to execute it. Here are a few examples:
Set HTTP encoding in the header:
Content-Type: text/xml, charset=utf-8
Use a meta tag in the HTML:
Set XML encoding in the first line of XML documents:
<?xml version="1.0" encoding="utf-8"?>
Remember, your AJAX application's attack surface is under your control. The choices you make can drastically affect the size of your application's attack surface. Be sure you ask questions about where your code runs, what data formats and protocols are involved, and which parsers and interpreters get invoked. And most importantly, be sure to nail down how you're going to keep code and data separate.®