Nothing about Collage, developed at Georgia Tech is particularly novel, as this Networkworld article, via Slashdot, explains. At its core it uses steganography, hiding messages in other information, a technique that the researchers themselves admit isn’t secure in and of itself. This is a well known limitation of steganography, it relies on obscurity rather than provably strong measures like the non-trivial math used in modern cryptography. It does sound like the hiding extends even to trying to make the network traffic used to pass messages blend into ordinary traffic one would expect to see at social networking sites and user content sharing sites. The use of a web site testing tool, Selenium, is pretty clever actually to achieve this end.
What Collage does rely on to get past censorship is an expectation that those trying to block communications would be unwilling to enact a massive blockade on the kinds of sites through which Collage operates. Ethan Zuckerman, at the Berkman Center, refers to this as the cute cat theory. I was skeptical of Twitter’s usage during the Iranian elections until my friend, Quinn Norton, referred me to this idea.
If Collage is able to adapt in the face of partial blocking, it would probably be all that much more effective to boot. Would censors back off if they thought they’d block enough to stifle speech? The more tools there are like this, the more options those needing to get around censorship have. I could also see this as part of a layered technique, swarming through open channels, obscured ones like Collage, and encrypted ones like Tor.