研究人员开发了一种新颖的越狱技术，用于对齐大型语言模型，该技术利用了同人小说子类型。该方法使用来自十二个不同Archive of Our Own (AO3) 子类型的段落来嵌入有害行为，绕过传统防御。该攻击将八个LLM的攻击成功率（ASR）从0.278显著提高到0.731，表明其有效性源于写作风格而非提示结构。提出的防御措施被发现无效，这表明需要转向基于语域的攻击。 AI
影响 这项研究突显了LLM安全训练中的新漏洞，可能需要超越简单提示过滤的新型防御机制。
排序理由 该集群包含一篇详细介绍越狱LLM新方法的论文。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2606.04483v1 Announce Type: new Abstract: Existing jailbreaks against aligned LLMs are discrete artifacts whose surface forms are easy to fingerprint and patch. We argue that the real failure mode is not any specific prompt, but an entire register of natural human writing t…
Existing jailbreaks against aligned LLMs are discrete artifacts whose surface forms are easy to fingerprint and patch. We argue that the real failure mode is not any specific prompt, but an entire register of natural human writing that safety training has under-covered. Building …