CoSafe：评估多轮对话共指消解中的大型语言模型安全性
CoSafe: Evaluating Large Language Model Safety in Multi-Turn Dialogue Coreference
AI总结：
本文首次研究多轮对话共指中的LLM安全性，构建了1,400个多轮共指攻击问题的CoSafe数据集，评估五个开源模型并揭示了其安全漏洞。
摘要：
随着大型语言模型（LLM）不断演进，确保其安全性仍是一个关键的研究问题。以往针对LLM安全的红队方法主要集中于单提示攻击或目标劫持。据我们所知，我们是首个研究多轮对话共指中LLM安全性的团队。我们创建了一个包含14个类别、共1,400个问题的数据集，每个问题均具有多轮共指安全攻击特征。随后，我们对五个广泛使用的开源LLM进行了详细评估。结果表明，在多轮共指安全攻击下，LLaMA2-Chat-7b模型的攻击成功率最高，为56%；而Mistral-7B-Instruct模型最低，为13.9%。这些发现凸显了LLM在对话共指交互中的安全漏洞。
英文摘要：
As large language models (LLMs) constantly evolve, ensuring their safety remains a critical research problem. Previous red-teaming approaches for LLM safety have primarily focused on single prompt attacks or goal hijacking. To the best of our knowledge, we are the first to study LLM safety in multi-turn dialogue coreference. We created a dataset of 1,400 questions across 14 categories, each featuring multi-turn coreference safety attacks. We then conducted detailed evaluations on five widely used open-source LLMs. The results indicated that under multi-turn coreference safety attacks, the highest attack success rate was 56% with the LLaMA2-Chat-7b model, while the lowest was 13.9% with the Mistral-7B-Instruct model. These findings highlight the safety vulnerabilities in LLMs during dialogue coreference interactions.
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。
