DynaPhish

Official repository for "Knowledge Expansion and Counterfactual Interaction for Reference-Based Phishing Detection". Published in USENIX Security 2023.

DynaPD 6K Phishing Kits Online Version • DynaPD 6K Phishing Kits Source Code • WebInteraction Driver: MyXdriver • Paper • Website • Citation

Introduction

Existing reference-based phishing detectors:

• ❌ Rely on a limited reference list which cannot adapt to temporal (e.g. emerging cryptocurrency brands) and regional (e.g. local brands) interests
• ❌ Unable to address logo-less phishing webpages
• ❌ Use un-interactable benchmark datasets as the test environment

In this work, we propose a framework called DynaPhish, as a complementary module for all reference-based phishing detectors. Our contributions lie in three folds:

• ✅ We perform on-the-fly knowledge expansion of the reference list automatically. Meanwhile, we use the popularity-based validation mechanism to ensure the benignity of added reference.
• ✅ We are the first to introduce the behavioral intention, which makes phishing decisions via observing the suspicious behaviors during the login action
• ✅ We publish DynaPD, which includes 6K clean and live phishing kits that are safe and interactable. Download from here: DynaPD. Visit the online demo here: DynaPD Dataset Demo.

Framework

Dynaphish consists of the following steps:

• Step 1: Run the reference-based detector as normal.
• Step 2: If the detector cannot recognize the phishing target, run the Brand Knowledge Expansion module. It will take the domain or the logo from the webpage, and search for the relevant brand with Google search API and Google OCR API.
• Step 3: If a brand can be returned from the Brand Knowledge Expansion module, we will expand the reference list and re-run step 1.
• Step 4: If the Brand Knowledge Expansion fails, we will run Web Interaction, this will check whether the webpage exhibits any suspicious behaviors during login.
• Step 5: A phishing alarm will be raised if either the reference-based detector or the Web Interaction reports the webpage as phishing.

Project Structure

|_ knowledge_expansion: Knowledge Expansion Module
  |_ brand_knowledge_online.py: Knowledge Expansion Class
|_ field_study_logo2brand: testing scripts
  |_ configs_template.yaml: configuration file for the models
  |_ dynaphish_main.py: main script

Setup

Requirements

• CUDA 11

Implemented and tested on Ubuntu 16.04 and 20.04, CUDA 11.1, cuDNN 10.1. Should work on other Debian-based systems as well.

1. Install the required packages by

chmod +x setup.sh
export ENV_NAME="dynaphish" && ./setup.sh

This script will create a new conda environment called dynaphish.

2. Update the configuration file for the knowledge expansion module

chmod +x update_config.sh
./update_config.sh

3. Create a google cloud service account, set the billing details

   • Create a project, enable "Custom Search API", "Cloud Vision API"
   • For "Custom Search API", get the API Key and Search Engine ID following this guide.
   • Create a blank txt file in the directory "knowledge_expansion/api_key.txt", copy and paste your API Key and Search Engine ID into the txt file like the following:

    [YOUR_API_KEY]
    [YOUR_SEARCH_ENGINE_ID]
   

   • For "Cloud Vision API", download the JSON key following this guide, save the JSON file under "knowledge_expansion/discoverylabel.json"

4. The main script is field_study_logo2brand/dynaphish_main.py

conda activate dynaphish
python -m field_study_logo2brand.dynaphish_main --folder [folder_to_test, e.g. datasets/test_sites] 

Citation

If you find our work useful, please consider citing our paper :)

@inproceedings {291106,
    author = {Ruofan Liu and Yun Lin and Yifan Zhang and Penn Han Lee and Jin Song Dong},
    title = {Knowledge Expansion and Counterfactual Interaction for {Reference-Based} Phishing Detection},
    booktitle = {32nd USENIX Security Symposium (USENIX Security 23)},
    year = {2023},
    isbn = {978-1-939133-37-3},
    address = {Anaheim, CA},
    pages = {4139--4156},
    url = {https://www.usenix.org/conference/usenixsecurity23/presentation/liu-ruofan},
    publisher = {USENIX Association},
    month = aug,
}

Contacts

I you encounter any issues in code deployment, please reach us via Email or create an issue in the repository: liu.ruofan16@u.nus.edu, lin_yun@sjtu.edu.cn