This project seeks to automate the censorship circumvention process, training AI to quickly test and learn viable circumvention techniques.
Pro-Internet freedom researchers and tool developers have long engaged in a manual cat-and-mouse game with censoring regimes, with censorship evaders finding ways to confuse censors into thinking that traffic is acceptable, and censors patching their system to thwart such efforts. This project takes a drastic departure from the previously manual evade-detect cycle by creating and utilizing techniques to automate the discovery of censorship evasion strategies, developing artificial intelligence to train against real censors and automatically learn how to circumvent censorship. The Geneva tool has been tested in China, India, and Kazakhstan, finding “…dozens of ways to circumvent censorship by exploiting gaps in censors’ logic and finding bugs that…would have been virtually impossible for humans to find manually.”
The created algorithms will be capable of evading censorship of a wide range of network protocols while seeking to integrate the techniques into other already-deployed censorship evasion tools.
