Internet Censorship and Online Speech

Our research group has been studying Internet censorship and online speech for over two decades, developing both measurement techniques to understand how content is filtered worldwide and circumvention tools to help users access information freely. More recently, we have expanded our focus to include content moderation practices on online platforms and generative AI systems.

Censorship Circumvention

We have developed several systems that help users circumvent Internet censorship, beginning with Infranet (2002), one of the first systems designed to enable users to surreptitiously retrieve blocked content via cooperating Web servers distributed across the Internet.

Collage (2010) introduced a novel approach that hides messages in user-generated content hosted on sites that censors are unlikely to block, using steganography and deniable communication techniques.

Facade (2014) provides high-throughput, deniable censorship circumvention by hiding traffic in Web search queries, making it difficult for censors to distinguish circumvention traffic from normal browsing activity.

Deniable Liaisons (2014) enables covert communication through traffic analysis resistance, allowing users to communicate without revealing the existence of the communication channel.

Censorship Research

Censorship Measurement

Understanding censorship requires measuring it at scale, but measurement itself poses ethical and safety challenges. Our work has advanced both the techniques and ethics of censorship measurement.

Encore (2015) enables lightweight, large-scale measurement of Web censorship using cross-origin requests, allowing researchers to detect blocking from many vantage points without requiring software installation.

GFWeb (2024) provides comprehensive measurement of China's Great Firewall at scale, revealing the multi-layered filtering apparatus used to control Web access. Related work examines how the Great Firewall discovers hidden circumvention servers.

We have also developed techniques for automated detection and fingerprinting of censorship block pages, enabling researchers to identify censorship across different countries and ISPs, and explored the ethical concerns and safety considerations for censorship measurement.

Online Speech

Content Moderation

As online platforms have become central to public discourse, understanding how they moderate content has become increasingly important. Our research examines content moderation policies and their effects on users across traditional platforms and emerging AI systems.

We have conducted in-depth studies of online platforms' content moderation policies, analyzing how community guidelines shape online discourse across major social media platforms. Our work examines the gap between stated policies and actual enforcement practices.

With the rise of generative AI, we are studying content moderation in AI products, examining how systems like ChatGPT enforce content policies and how users experience and respond to these restrictions.

Our work also addresses disinformation, developing infrastructure-based techniques to identify disinformation websites using network and hosting features without analyzing content directly.

Selected Publications

Content Moderation

Censorship Measurement

Censorship Circumvention