使用低资源语言对LLM进行多语言劫持
Multilingual jailbreaking of LLMs using low-resource languages
AI总结：
研究通过使用低资源非洲语言进行多轮对话来测试大型语言模型的安全机制，发现翻译质量是影响低资源语言劫持成功率的关键因素。
摘要：
大型语言模型（LLMs）仍然容易受到绕过安全防护措施的劫持攻击。我们研究了使用低资源非洲语言（阿弗里卡语、基索瓦希利、isiXhosa和isiZulu）的多轮对话是否能绕过商业LLM的安全机制。我们翻译了现有数据集中的提示，并通过自动化测试和本地母语者的人工红队测试评估了ChatGPT、Claude、DeepSeek、Gemini和Grok。单轮翻译攻击效果不佳，而多轮对话在英语有害响应率方面从52.7%（Claude 3.5 Haiku）到83.6%（GPT-4o-mini），阿弗里卡语从60.0%（Claude 3.5 Haiku）到78.2%（GPT-4o-mini），基索瓦希利从41.8%（Claude 3.5 Haiku）到70.9%（DeepSeek）。人工红队测试比自动化方法提高了劫持率。所有评估语言的平均劫持率从59.8%增加到75.8%，其中阿弗里卡语提高了20.0%，isiZulu提高了12.7%，isiXhosa提高了12.3%，基索瓦希利提高了1%，这表明翻译质量限制了劫持的成功率。这些发现表明，LLM中的漏洞在多语言环境中仍然存在，翻译质量是决定低资源语言劫持成功率的关键因素。
英文摘要：
Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails. We investigate whether multi-turn conversations using low-resource African languages (Afrikaans, Kiswahili, isiXhosa, and isiZulu) can bypass safety mechanisms across commercial LLMs. We translated prompts from existing datasets and evaluated ChatGPT, Claude, DeepSeek, Gemini, and Grok through automated testing and human red-teaming with native speakers. Single-turn translation attacks proved ineffective, while multi-turn conversations achieved English harmful response rates from 52.7% (Claude 3.5 Haiku) to 83.6% (GPT-4o-mini), Afrikaans from 60.0% (Claude 3.5 Haiku) to 78.2% (GPT-4o-mini), and Kiswahili from 41.8% (Claude 3.5 Haiku) to 70.9% (DeepSeek). Human red-teaming increased jailbreak rates compared to automated methods. Over all evaluated languages, the average jailbreak rate increased from 59.8% to 75.8%, with improvements of +20.0% (Afrikaans), +12.7% (isiZulu), +12.3% (isiXhosa), and +1% (Kiswahili), demonstrating that poor translation quality limits jailbreak success. These findings suggest that vulnerabilities in LLMs persist in multilingual contexts and that translation quality is the critical factor determining jailbreak success in low-resource languages.
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。
