English(EN) AttentionSwitch: Jailbreak Defense for LLMs via Joint Learning of Sparse Attention-Head Selection and Scaling

新的AttentionSwitch防御机制联合优化LLM安全干预

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-09 05:00

研究人员开发了AttentionSwitch，这是一种新颖的防御机制，旨在保护大型语言模型（LLM）免受越狱攻击。与先前微调模型或顺序优化干预的方法不同，AttentionSwitch联合学习选择特定的注意力头并应用查询条件缩放因子。这种方法可以在不改变其表示方向的情况下衰减或放大选定的注意力头，从而在保持LLM原始能力的同时提高对恶意输入的安全性。评估表明，AttentionSwitch的性能优于现有防御方法，尤其是在训练数据有限的情况下。 AI

影响通过提供一种更有效、更高效的防御越狱攻击的方法，同时不损害模型能力，从而增强了LLM的安全性。

排序理由这是一篇详细介绍LLM安全新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 16 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 这是一篇详细介绍LLM安全新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Doniyorkhon Obidov, Honggang Yu, Xiaolong Guo, Kaichen Yang · 2026-11-09 05:00

AttentionSwitch：通过稀疏注意力头选择与缩放的联合学习实现大语言模型的越狱防御

arXiv:2611.08752v1 Announce Type: cross Abstract: Large language models have achieved remarkable performance across a wide range of tasks, yet they remain susceptible to jailbreak attacks that circumvent safety mechanisms and elicit harmful or unintended outputs. Although safety …

本文如何被排名

报道来源 [1]

AttentionSwitch：通过稀疏注意力头选择与缩放的联合学习实现大语言模型的越狱防御

相关实体

相关话题