Backdoor-Enhanced-Alignment

This is the official code repository for the paper Mitigating Fine-tuning Jailbreak Attack with Backdoor Enhanced Alignment.

For now, we only realease a demo example for GPT-3.5 experiments through OpenAI API. The complete evaluation pipelines and support for more open source LLMs will be released soon.

Citation

Please cite the following preprint when referencing our paper:

@misc{wang2024mitigating,
        title={Mitigating Fine-tuning Jailbreak Attack with Backdoor Enhanced Alignment}, 
        author={Jiongxiao Wang and Jiazhao Li and Yiquan Li and Xiangyu Qi and Junjie Hu and Yixuan Li and Patrick McDaniel and Muhao Chen and Bo Li and Chaowei Xiao},
        year={2024},
        eprint={2402.14968},
        archivePrefix={arXiv},
        primaryClass={cs.CR}
}