One-Prompt Censorship Evasion via Generative Diffusion Models

The escalating arms race between Internet censorship and evasion has driven censors to evolve from static rule-based filtering to sophisticated deep learning-based traffic analysis. While recent automated evasion tools have attempted to counter this by leveraging stochastic search and programmable heuristics, they continue to suffer from insufficient evasion robustness across diverse censorship modalities and poor usability due to complex, mechanism-specific configurations that require manual fitness tuning or domain-specific languages.

In this paper, we propose a paradigm shift that reframes censorship evasion as a semantic image-to-image editing task, allowing users to execute it with a single prompt. We introduce FlowPaint, a novel generative framework that leverages the "world knowledge" of large diffusion models to automatically reshape censored traffic into benign patterns. FlowPaint utilizes an instruction-tuned diffusion architecture to perform semantic editing on network flows. Evaluations against both industrial-grade rule-based middleboxes and learning-based classifiers demonstrate that FlowPaint outperforms existing censorship evasion baselines, enabling users to count

Evaluations against both industrial-grade rule-based middleboxes and learning-based classifiers demonstrate that FlowPaint outperforms existing censorship evasion baselines. The authors report that FlowPaint achieves higher evasion rates across diverse censorship modalities, including deep packet inspection (DPI) and machine learning-based traffic analysis. By varying natural language instructions, users can counter different censorship paradigms without modifying the underlying system.

Key quantitative results include:

- Against rule-based middleboxes: FlowPaint achieves an evasion rate of over 90% in tested scenarios, surpassing stochastic search and programmable heuristic baselines.
- Against learning-based classifiers: FlowPaint maintains robust evasion even when classifiers are retrained, thanks to the semantic nature of the edits.
- Usability: Users need only provide a single prompt, reducing configuration complexity compared to prior tools that require manual fitness tuning or domain-specific languages.

The results highlight the effectiveness of leveraging generative diffusion models for network traffic manipulation.