Measuring DNS Censorship of Generative AI Platforms

TL;DR

This work addresses how Generative AI platforms are censored via DNS by conducting large-scale DNS probing across approximately 850k resolvers to detect domain-level blocking. The authors assemble a diverse set of 61 Generative AI domains and measure DNS responses over about two months, revealing China as the dominant DNS censor with nationwide domain blocking and Russia that shows selective, ASN-based censorship. They provide a cross-platform comparison with OONI, Censored Planet, and GFWatch, highlighting that existing platforms offer limited Generative AI-domain granularity. The study contributes the first DNS-focused view of Generative AI censorship, exposes distinct country-specific blocking patterns and methods, and outlines limitations and directions for longer-term, multi-protocol measurements.

Abstract

Generative AI is an invaluable tool, however, in some parts of the world, this technology is censored due to political or societal issues. In this work, we monitor Generative AI censorship through the DNS protocol. We find China to be a leading country of Generative AI censorship. Interestingly, China does not censor all AI domain names. We also report censorship in Russia and find inconsistencies in their process. We compare our results to other measurement platforms (OONI, Censored Planet, GFWatch), and present their lack of data on Generative AI domains.