Proteus

This repository accompanies our NDSS 2026 paper: Enhancing Website Fingerprinting Attacks against Traffic Drift.

Proteus introduces an adaptive fine-tuning framework that significantly enhances the robustness of website fingerprinting (WF) attacks against temporal, version, network, behavioral, and open-world traffic drift scenarios, by leveraging unlabeled traffic for continual adaptation.

---

📂 Datasets

We collected over 350,000 real-world Tor traffic traces, covering diverse drift scenarios.

• The dataset is organized into six categories corresponding to the experimental settings in our paper.
• Download the datasets via link.

mkdir datasets

Extract all datasets and place them under the datasets/ directory.

---

⚙️ Installation

Clone this repository and install the dependencies:

cd wflib_copy
pip install --user .
cd ..

More details can be found in WFlib.

---

🧪 Running Experiments

We provide scripts to reproduce the main experiments reported in our NDSS 2026 paper.

• Temporal Drift

bash scripts/TemporalDrift/Proteus.sh

• Version Drift

bash scripts/VersionDrift/Proteus.sh

• Network Drift

bash scripts/NetworkDrift/Proteus.sh

• Behavioral Drift

bash scripts/BehaviorDrift/Proteus.sh

• Open-World Setting

bash scripts/OpenWorld/Proteus.sh

• Against Defenses

bash scripts/Defense/Proteus.sh

---

📖 Citation

If you use this code or dataset in your research, please cite our paper:

@inproceedings{proteus2026,
  title     = {Enhancing Website Fingerprinting Attacks against Traffic Drift},
  author    = {Xinhao Deng, Yixiang Zhang, Qi Li, Zhuotao Liu, Yabo Wang, Ke Xu},
  booktitle = {Network and Distributed System Security (NDSS) Symposium},
  year      = {2026}
}

---

📜 License

This project is released under the MIT License. See LICENSE for details.