A Trick That Hides Censored Websites Inside Cat Videos

A pair of researchers behind a system for avoiding internet censorship wants to deliver banned websites inside of cat videos. Their system uses media from popular, innocuous websites the way a high schooler might use the dust jacket of a textbook to hide the fact that he’s reading a comic book in class. To the overseeing authority—in the classroom, the teacher; on the internet, a government censor—the content being consumed appears acceptable, even when it’s illicit.

The researchers, who work at the University of Waterloo’s cryptography lab, named Slitheen after a race of aliens from Doctor Who who wear the skins of their human victims to blend in. The system uses a technique called decoy routing, which allows users to view blocked sites—like a social-networking site or a news site—while generating a browsing trail that looks exactly as if they were just browsing for shoes or watching silly videos on YouTube.

Slitheen’s web browser starts the process by sending off a normal request for a harmless, “overt” site—but in it, it embeds a secret tag: an encrypted, second request for the user’s true target, a sensitive “covert” website. Website requests pass through relay stations built into the internet’s infrastructure on their way from a browser to a web server. If a relay station were to install Slitheen software, when the request passes through it, the station would detect the secret request and decode it using its secret key. (A non-Slitheen relay station that doesn’t have the right secret key wouldn’t even be able to tell that there is a secret request bundled inside of the traffic, let alone decrypt it.)

Once the relay has decrypted the secret request, it begins the process of assembling the unique package that defines the Slitheen technique. It downloads the overt site, which will act as the camouflage for the sensitive data, and simultaneously downloads the target covert site—the payload that the user really wants to see.

It then strips out all the images and videos from the overt site, and replaces them with the entire contents of the covert site. If the whole covert site is smaller than the video and image data from the overt site, the relay station will add junk data until the sizes are identical. If the covert site is bigger, it can be split across multiple overt requests to different safe websites.

Finally, once the covert site has been hidden inside the overt site, the relay sends the whole bundle back to the user that requested it. When it arrives, the user’s Slitheen client sifts through the data, loading the overt site in the background, and displaying the covert site in a browser.

Slitheen’s defining feature is that the complex traffic it generates is indistinguishable from a normal request. That is, two computers sitting next to one another, downloading data from Amazon.com’s homepage—one that does so normally and another with the contents of this Atlantic story instead of Amazon’s images and videos—would create identical traffic patterns. The more complex Slitheen request would take slightly longer to come back, but its defining characteristics, from packet size to timing, would be the same.

To avoid being caught, it’s important to conform decoy traffic to normal traffic as closely as possible. Decoy routing was first proposed in 2011, but it was quickly discovered to be vulnerable to fingerprinting attacks, which involve analyzing patterns as encrypted data moves across the internet. Since the data is encrypted, censors can’t peer directly into the data stream to see what’s being downloaded. But if they know what traffic to a banned site looks like—how fast it downloads, and in what order different-sized chunks of data arrive—it can identify when that site’s being accessed just based on the data streams that pass through the networks it surveils. Last month, I wrote about a technique that uses fingerprinting to guess what YouTube video is being watched.

Slitheen is designed to work even if a censor knows exactly where the relay stations are, how they work, and which sites are being used as overt sites. “They still should not be able to distinguish a Slitheen connection from a normal connection,” said Ian Goldberg, a computer-science professor at Waterloo and one of Slitheen’s developers. Only if the censor obtained Slitheen’s private key—a randomly generated 256-bit number—could it decode the secret requests embedded within the overt ones. But that secret is easily protected, and can be swapped for a new one at a moment’s notice.

Slitheen’s technical details are sound, said Gustaf Björksten, the chief technologist at Access Now, a digital human-rights organization. But Björksten, who wasn’t involved in developing the system, worried that some thorny practical obstacles could make it hard to widely implement a decoy routing system such as this.

Slitheen needs participation from friendly internet providers in order to function. But the system requires more sophisticated hardware and software than are usually used for internet routers, said Cecylia Bocovich, a graduate student at Waterloo and the primary author of the paper introducing Slitheen. This presents the obstacle of convincing providers to set up complex decoy relay stations out of the goodness of their hearts.

Indeed, even though decoy routing was first presented more than five years ago, it hasn’t been widely adopted. But Bocovich says she and Goldberg are in talks to deploy some Slitheen infrastructure within a year. They wouldn’t share any details about the pending deployment.

Björksten also worried that the developers’ decision to prioritize security—thus sacrificing some speed—could drive away users who are unwilling to use a system that inhibits their browsing habits. In places without extensive infrastructure for Tor, a network of servers that allows for anonymous browsing, many users abandon the service because it’s too slow, Björksten said. The same problem could befall Slitheen, or any censorship circumvention technology, if it causes slowdowns. But for now, there aren’t any good alternatives: Security gains come with usability drawbacks.

Bocovich and Goldberg say they’re working on a few improvements before Slitheen is deployed in the real world. One of them will make Slitheen’s overt connections “look more like humans are using the sites,” Bocovich said. Instead of downloading the same site over and over, or jumping between sites on a roster of popular destinations, it could try to follow links in a way that mimics genuine browsing patterns.

But that extra wrinkle in an already complex system should have no effect on a Slitheen user. It will only change the pattern of fake traffic a nosy censor would see on a network. Behind their computers, users will continue leaving a trail of unremarkable browsing histories, as they enjoy unfiltered access to the internet.