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 2502.15086cs.CL
收藏

安全标准对每个人都一样吗?大语言模型的用户特定安全评估

Is Safety Standard Same for Everyone? User-Specific Safety Evaluation of Large Language Models

KAIST(韩国科学技术院);Adobe Research(Adobe 研究院)

Yeonjun In, Wonjoong Kim, Kanghoon Yoon, Sungchul Kim, Mehrab Tanjim, Sangwu Park, Kibum Kim, Chanyoung Park

AI总结:

针对现有LLM安全评估忽视用户特定标准的问题,提出U-SafeBench基准,评估20个LLM发现其均未满足用户特定安全要求,并基于思维链提出简单补救措施以提升安全性。

摘要:

随着大语言模型(LLM)智能体的使用持续增长,其安全漏洞也日益明显。大量基准测试通过严重依赖通用标准来定义安全性,从而评估LLM安全的各个方面,却忽视了用户特定的标准。然而,LLM的安全标准可能因用户特定画像而异,而非在所有用户中普遍一致。这引发了一个关键的研究问题:在考虑用户特定安全标准时,LLM智能体是否能够安全行事?尽管这对LLM的安全使用至关重要,但目前尚无基准数据集来评估LLM的用户特定安全性。为弥补这一空白,我们引入了U-SafeBench,一个旨在评估LLM安全中用户特定方面的基准。我们对20个广泛使用的LLM的评估显示,当前LLM在考虑用户特定安全标准时未能安全行事,这标志着该领域的一个新发现。为解决这一漏洞,我们提出了一种基于思维链的简单补救措施,并证明了其在提高用户特定安全性方面的有效性。我们的基准和代码可在https://github.com/yeonjun-in/U-SafeBench获取。

英文摘要:

As the use of large language model (LLM) agents continues to grow, their safety vulnerabilities have become increasingly evident. Extensive benchmarks evaluate various aspects of LLM safety by defining the safety relying heavily on general standards, overlooking user-specific standards. However, safety standards for LLM may vary based on a user-specific profiles rather than being universally consistent across all users. This raises a critical research question: Do LLM agents act safely when considering user-specific safety standards? Despite its importance for safe LLM use, no benchmark datasets currently exist to evaluate the user-specific safety of LLMs. To address this gap, we introduce U-SafeBench, a benchmark designed to assess user-specific aspect of LLM safety. Our evaluation of 20 widely used LLMs reveals current LLMs fail to act safely when considering user-specific safety standards, marking a new discovery in this field. To address this vulnerability, we propose a simple remedy based on chain-of-thought, demonstrating its effectiveness in improving user-specific safety. Our benchmark and code are available at https://github.com/yeonjun-in/U-SafeBench.

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。

↑