arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2404.01318cs.CRcs.LG
收藏

JailbreakBench:面向大语言模型越狱攻击的开放鲁棒性基准

JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models

University of Pennsylvania(宾夕法尼亚大学);ETH Zurich(苏黎世联邦理工学院);EPFL(洛桑联邦理工学院);Sony AI(索尼人工智能)
Patrick Chao, Edoardo Debenedetti, Alexander Robey, Maksym Andriushchenko, Francesco Croce, Vikash Sehwag, Edgar Dobriban, Nicolas Flammarion, George J. Pappas,… 展开作者

Patrick Chao, Edoardo Debenedetti, Alexander Robey, Maksym Andriushchenko, Francesco Croce, Vikash Sehwag, Edgar Dobriban, Nicolas Flammarion, George J. Pappas, Florian Tramer, Hamed Hassani, Eric Wong

AI总结:

针对当前大语言模型越狱评估缺乏统一标准、结果不可比且难复现的问题,研究团队推出开源基准JailbreakBench,整合越狱构件库、合规数据集、标准化评估框架与排行榜,助力越狱攻击与防御的可复现研究。

摘要:

越狱攻击会导致大语言模型(LLM)生成有害、不道德或其他违规内容。评估这类攻击存在诸多挑战,当前的基准集合与评估技术未能充分应对这些问题。首先,越狱评估尚无明确的行业实践标准。其次,现有研究以不可相互比较的方式计算成本与成功率。第三,大量研究结果无法复现,原因在于它们不公开对抗性提示、采用闭源代码,或是依赖持续迭代的专有API。为解决这些挑战,我们推出JailbreakBench,这一开源基准包含以下组成部分:(1)一个持续更新的前沿对抗性提示资源库,我们将其称为越狱构件;(2)一个包含100种行为的越狱数据集,行为既有原创的,也来自过往研究(Zou等人,2023;Mazeika等人,2023、2024),且均符合OpenAI的使用政策;(3)一个部署于https://github.com/JailbreakBench/jailbreakbench的标准化评估框架,包含明确定义的威胁模型、系统提示、对话模板与评分函数;(4)一个部署于https://jailbreakbench.github.io/的排行榜,追踪各类大语言模型上攻击与防御方法的性能。我们已审慎考量发布该基准的潜在伦理影响,认为它对研究社区整体而言利大于弊。

英文摘要:

Jailbreak attacks cause large language models (LLMs) to generate harmful, unethical, or otherwise objectionable content. Evaluating these attacks presents a number of challenges, which the current collection of benchmarks and evaluation techniques do not adequately address. First, there is no clear standard of practice regarding jailbreaking evaluation. Second, existing works compute costs and success rates in incomparable ways. And third, numerous works are not reproducible, as they withhold adversarial prompts, involve closed-source code, or rely on evolving proprietary APIs. To address these challenges, we introduce JailbreakBench, an open-sourced benchmark with the following components: (1) an evolving repository of state-of-the-art adversarial prompts, which we refer to as jailbreak artifacts; (2) a jailbreaking dataset comprising 100 behaviors -- both original and sourced from prior work (Zou et al., 2023; Mazeika et al., 2023, 2024) -- which align with OpenAI's usage policies; (3) a standardized evaluation framework at https://github.com/JailbreakBench/jailbreakbench that includes a clearly defined threat model, system prompts, chat templates, and scoring functions; and (4) a leaderboard at https://jailbreakbench.github.io/ that tracks the performance of attacks and defenses for various LLMs. We have carefully considered the potential ethical implications of releasing this benchmark, and believe that it will be a net positive for the community.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑