Autosonda: Discovering Rules and Triggers of Censorship Devices

Using censorship to forbid access to certain content on the internet is very common in the world today. Some censorship mechanisms are well-studied, however there remain a large number of techniques that remain unknown. Furthermore, many censorship implementations are dynamic as they attempt to prevent against new circumvention techniques. Current tools often tell us when something is censored, but don’t do an automated analysis of the approach nor provide clues about the rule sets used by censorship devices. This paper presents Autosonda, a tool for discovering and studying decision models of censorship devices. Through network traffic alone, Autosonda fingerprints censorship devices by discovering their models and mechanisms for how they enforce rule sets. The strength of Autosonda is demonstrated in a study that we present of 76 web filters currently in use in the New York City metropolitan area. In our study we encounter a great variety of behavior and implementation techniques for blocking prohibited web content. Not only does Autosonda help us to find implementation flaws and rule sets, it also allows us to find circumvention paths for 100% of our test subjects. Being able to perform this type of detailed analysis automatically and at scale is a large contribution for understanding censorship and how device behavior can be classified.