@inproceedings{chakraborty-etal-2024-textual,
    title = "Can Textual Unlearning Solve Cross-Modality Safety Alignment?",
    author = "Chakraborty, Trishna  and
      Shayegani, Erfan  and
      Cai, Zikui  and
      Abu-Ghazaleh, Nael B.  and
      Asif, M. Salman  and
      Dong, Yue  and
      Roy-Chowdhury, Amit  and
      Song, Chengyu",
    editor = "Al-Onaizan, Yaser  and
      Bansal, Mohit  and
      Chen, Yun-Nung",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2024",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/add-emnlp-2024-awards/2024.findings-emnlp.574/",
    doi = "10.18653/v1/2024.findings-emnlp.574",
    pages = "9830--9844",
    abstract = "Recent studies reveal that integrating new modalities into large language models (LLMs), such as vision-language models (VLMs), creates a new attack surface that bypasses existing safety training techniques like supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF). While further SFT and RLHF-based safety training can be conducted in multi-modal settings, collecting multi-modal training datasets poses a significant challenge. Inspired by the structural design of recent multi-modal models, where all input modalities are ultimately fused into the language space, we explore whether unlearning solely in the textual domain can be effective for cross-modality safety alignment. Our empirical evaluation across seven datasets demonstrates promising transferability {---} textual unlearning in VLMs significantly reduces the Attack Success Rate (ASR) to less than 8{\%} and in some cases, even as low as nearly 2{\%} for both text-based and vision-text-based attacks, alongside preserving the utility. Moreover, our experiments show that unlearning with a multi-modal dataset offers no potential benefits but incurs significantly increased computational demands."
}