ConvoSentinel

This is the official repository of the EMNLP 2024 paper: Defending Against Social Engineering Attacks in the Age of LLMs.

SEConvo Data

The SEConvo data is available for download here. See SEConvo/README.md for data details.

To Run ConvoSentinel

1. Recommended to use python 3.10.
2. Obtain the trained_si_model and faiss_index_v5 (conversation snippet dataset). Place trained_si_model in the message_si_detector/ folder and faiss_index_v5 in the se_attempt_detector/ folder. These models are publicly available but are not detailed here due to the anonymity policy.
3. Place your Hugging Face token in the hf_token.txt file and your OpenAI API key in the openai_api_key.txt file.
4. Install the required packages:

   pip install -r requirements.txt

5. Run the ConvoSentinel script:

   bash run_convosentinel.sh SEConvo/annotated_test.json outputs/message_si_preds.json outputs/message_se_preds.json outputs/conv_se_preds.json

   This script generates:
   • Message-level SI predictions: outputs/message_si_preds.json
   • Message-level SE predictions: outputs/message_se_preds.json
   • Conversation-level SE predictions: outputs/conv_se_preds.json
6. By default, the script runs with cuda:0. Modify this in the script if necessary.

If you find our method useful, please cite us:

@article{ai2024defending,
  title={Defending Against Social Engineering Attacks in the Age of LLMs},
  author={Ai, Lin and Kumarage, Tharindu and Bhattacharjee, Amrita and Liu, Zizhou and Hui, Zheng and Davinroy, Michael and Cook, James and Cassani, Laura and Trapeznikov, Kirill and Kirchner, Matthias and others},
  journal={arXiv preprint arXiv:2406.12263},
  year={2024}
}