My friend and colleague Hal Roberts is one of the internet’s top censorship and filtering researchers. When Chinese authorities announced that a client-based piece of filtering software called GreenDam would be required for installation on new PCs, Hal downloaded the software and spent a good chunk of this week trying to understand how it works. He and colleagues released a report earlier today that demonstrates that GreenDam is incredibly ambitious and invasive to user privacy, but also badly broken and virtually unusable.
Perhaps the best demonstration of this fact is the video Hal put together earlier today and posted on his blog. It shows GreenDam slowly realizing that it wants to block a Falun Gong site… then blocking any other sites that begin with the letter “f”, due to the way Internet Explorer handles autocompletion and GreenDam blocks content.
Basically, GreenDam approaches censorship very differently from conventional filtering tools. Most client-side filters rely on lists of banned sites downloaded from the internet. Not GreenDam – Hal’s test shows that it doesn’t appear to know that falundafa.org (a site that the Great Firewall always blocks) is a banned site until he comes across a sensitive phrase, three clicks into the site. That phrase is checked against an internal dictionary. This allows GreenDam not only to censor the net, but to shut down an application like Notepad if someone writes sensitive content. Similarly GreenDam apparently doesn’t keep watchlists of porn sites – instead, it attempts to detect if an image is pornographic by checking against an internal algorithm. (Some testers have discovered that the algorithm blocks light-skinned naked people, though not dark-skinned ones.)
My guess is that problems like the ones Hal documents in his report and video are going to prevent GreenDam from being mandated across China. And I have confidence that Chinese hackers will find effective ways to shut off the tool shortly after it’s released widely. But it’s hard to know… and the ambitions of the tool designers to censor the ability to even type certain phrases is pretty chilling.