CAREER: Automatically Learning to Evade Internet Censorship

点击翻译按钮获取中文摘要

The Internet provides unprecedented opportunities for open communication, diplomacy, and education. Unfortunately, the openness of the Internet is challenged by powerful countries around the world who today engage in nationwide censorship of Internet traffic. For decades, security researchers have engaged in a cat-and-mouse game with censors, developing new schemes to evade censors, who in turn have developed increasingly sophisticated countermeasures. Censors have long had an inherent advantage in this arms race: details of their systems are typically not made publicly known, and thus researchers have had to undergo manual, laborious steps of measuring, innovating, implementing, and testing for new evasion techniques. This project proposes an ambitious research agenda towards developing artificial intelligence to automate the rapid discovery of new methods for evading and understanding nation-state censors. The ultimate goal of the project is to safely reach the logical conclusion of the evade/detect arms race--and to prepare for the next one. This project also includes an education plan that seeks to address the meteoric rise of enrollment in undergraduate computer science programs, by exploring ways to scale-up and broaden participation in undergraduate research. The project proposes Breakerspace, a lab designed around group undergraduate research projects, and integrates these into the work on automating censorship evasion.The proposed research follows three broad thrusts: (1) Developing new AI-based techniques for automatically evading in-network censors of various kinds, (2) Deploying AI-assisted censorship evasion strategies and developing new algorithms to efficiently scale-up discovery of new strategies via collaborative, crowd-sourced training, and (3) Performing AI-assisted measurement of censorship at unprecedented scale. The proposed research plan takes a practical approach--training, evaluating, measuring, and deploying against real nation-state censors, and making evasion software freely available for censored users. If successful, the AI this project builds and deploys, and the measurements it performs, will enable more agile evasion of new forms of censorship, and will lend deeper insight into how censors (fail to) work, how to circumvent them, and how they update over time. As a result, this project has the potential to break the manual evade/detect cycle that researchers and censoring regimes have engaged in for decades, thereby assisting millions of users around the world in achieving open access to information.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.