Unleashing the Unseen: Harnessing Benign Datasets for Jailbreaking Large Language Models
保存先:
| 出版年: | arXiv.org (Dec 19, 2024), p. n/a |
|---|---|
| 第一著者: | |
| その他の著者: | , , |
| 出版事項: |
Cornell University Library, arXiv.org
|
| 主題: | |
| オンライン･アクセス: | Citation/Abstract Full text outside of ProQuest |
| タグ: |
タグなし, このレコードへの初めてのタグを付けませんか!
|
MARC
| LEADER | 00000nab a2200000uu 4500 | ||
|---|---|---|---|
| 001 | 3147571297 | ||
| 003 | UK-CbPIL | ||
| 022 | |a 2331-8422 | ||
| 035 | |a 3147571297 | ||
| 045 | 0 | |b d20241219 | |
| 100 | 1 | |a Zhao, Wei | |
| 245 | 1 | |a Unleashing the Unseen: Harnessing Benign Datasets for Jailbreaking Large Language Models | |
| 260 | |b Cornell University Library, arXiv.org |c Dec 19, 2024 | ||
| 513 | |a Working Paper | ||
| 520 | 3 | |a Despite significant ongoing efforts in safety alignment, large language models (LLMs) such as GPT-4 and LLaMA 3 remain vulnerable to jailbreak attacks that can induce harmful behaviors, including through the use of adversarial suffixes. Building on prior research, we hypothesize that these adversarial suffixes are not mere bugs but may represent features that can dominate the LLM's behavior. To evaluate this hypothesis, we conduct several experiments. First, we demonstrate that benign features can be effectively made to function as adversarial suffixes, i.e., we develop a feature extraction method to extract sample-agnostic features from benign dataset in the form of suffixes and show that these suffixes may effectively compromise safety alignment. Second, we show that adversarial suffixes generated from jailbreak attacks may contain meaningful features, i.e., appending the same suffix to different prompts results in responses exhibiting specific characteristics. Third, we show that such benign-yet-safety-compromising features can be easily introduced through fine-tuning using only benign datasets. As a result, we are able to completely eliminate GPT's safety alignment in a blackbox setting through finetuning with only benign data. Our code and data is available at \url{https://github.com/suffix-maybe-feature/adver-suffix-maybe-features}. | |
| 653 | |a Feature extraction | ||
| 653 | |a Datasets | ||
| 653 | |a Alignment | ||
| 653 | |a Large language models | ||
| 700 | 1 | |a Li, Zhe | |
| 700 | 1 | |a Li, Yige | |
| 700 | 1 | |a Sun, Jun | |
| 773 | 0 | |t arXiv.org |g (Dec 19, 2024), p. n/a | |
| 786 | 0 | |d ProQuest |t Engineering Database | |
| 856 | 4 | 1 | |3 Citation/Abstract |u https://www.proquest.com/docview/3147571297/abstract/embedded/ZKJTFFSVAI7CB62C?source=fedsrch |
| 856 | 4 | 0 | |3 Full text outside of ProQuest |u http://arxiv.org/abs/2410.00451 |