arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2602.10382cs.CL
收藏

触发器劫持语言电路:大型语言模型中后门行为的机制分析

Language Triggers Hijack Language Circuits: A Mechanistic Analysis of Backdoor Behaviors in Large Language Models

Inria Paris(巴黎研究所)

Théo Lasnier, Wissam Antoun, Francis Kulumba, Benoît Sagot, Djamé Seddah

AI总结:

本文研究了大型语言模型中后门行为的机制,发现触发器通过劫持现有语言组件实现输出语言切换,为后门防御提供了新思路。

摘要:

后门攻击对大型语言模型(LLMs)构成了显著的安全风险,然而触发器在内部机制上如何运作仍然知之甚少。我们首次对语言切换后门进行了机制分析,研究了包含在预训练过程中注入触发器的GAPperon模型家族(1B、8B、24B参数),这些触发器导致输出语言切换。通过激活修补技术,我们确定触发器的形成集中在早期层(模型深度的7.5-25%),并识别出哪些注意力头处理触发信息。我们的核心发现是,激活触发器的头在不同模型规模中与自然编码输出语言的头有显著重叠,重叠指数在0.18到0.66之间。这表明后门触发器并不形成孤立的电路,而是利用模型现有的语言组件。这些发现对后门防御有影响:检测方法可能受益于监控已知的功能组件而非搜索隐藏的电路,缓解策略可能可以利用注入行为与自然行为之间的纠缠。

英文摘要:

Backdoor attacks pose significant security risks for Large Language Models (LLMs), yet the internal mechanisms by which triggers operate remain poorly understood. We present the first mechanistic analysis of trigger-induced language-switching backdoors injected during pre-training, studying the Gaperon model family (1B, 8B and 24B). Using activation patching, we localize trigger formation and identify which attention heads process trigger and natural language information. Our central finding is that trigger heads substantially overlap with heads naturally encoding output language across model scales, with Jaccard indices between 0.18 and 0.43 over the top 10 heads identified. This suggests that backdoor triggers do not form new circuits but instead co-opt the model's existing language components and representations. These findings have implications for backdoor defense as detection methods and mitigation strategies could leverage this entanglement between triggers and natural behaviors. More broadly, our work represents a first step toward a more realistic mechanistic understanding of pre-training-injected backdoors in LLMs, paving the way for principled, interpretability-driven defenses.

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。

↑