English(EN) Towards Merging-conditional Backdoor Unalignment against Large Language Models

模型合并中发现新的LLM后门漏洞

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-15 04:00

研究人员发现了一种名为“单向合并条件后门”（OMCB）的大型语言模型（LLM）新安全漏洞。该漏洞在单个模型中处于休眠状态，但在模型合并时会被激活，导致安全不对齐。一种名为SMOKE的新攻击框架采用元学习方法，通过模拟合并过程植入这些后门，即使对现有的后门检测方法也有效。 AI

影响突出了LLM合并中一个关键的、先前被忽视的漏洞，可能危及安全对齐。

排序理由详细介绍LLM新安全漏洞和攻击方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

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; 详细介绍LLM新安全漏洞和攻击方法的学术论文。[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。

arXiv cs.AI TIER_1 English(EN) · Yukun Chen, Yidou Chen, Yiming Li, Xinyu Zhang, Baosong Yang, Hongwei Yao, Zhan Qin · 2026-10-15 04:00

面向大型语言模型的合并条件后门不对齐研究

arXiv:2610.14701v1 Announce Type: cross Abstract: Model merging has become widely adopted for enhancing the task-specific performance of Large Language Models (LLMs) without additional training. While the specialized model from third-party platforms may undergo safety audits (e.g…