ACL Findings2026
Neuronal Insights into LLM Attacks: Targeted Neuron Tuning for Precise and Robust Vulnerability Patching
Dan Shi, Renren Jin, Zhuowen Han, Yuqi Ren, Xinwei Wu, Zhigen Li, Deyi Xiong
Abstract
Despite recent advances in safety alignment, large language models (LLMs) remain highly susceptible to adversarial attacks, while the internal mechanisms behind such vulnerabilities are still poorly understood. Existing gradientbased attribution methods offer valuable interpretability for analyzing information storage and processing in LLMs. However, they are inapplicable to adversarial attacks, which typically occur in open-ended generation settings without fixed ground-truth outputs. To address these challenges, we propose a novel similaritybased gradient attribution method to identify key neurons sensitive to adversarial behaviors in open-ended generation tasks. The detected neurons, termed targeted neurons, play a critical role in safety training. Building on this neuron-level perspective, we uncover two key neuronal patterns: (i) universal neurons that are consistently exploited across multiple attack strategies, and (ii) interference neurons that hinder safety improvements when fine-tuned indiscriminately, providing mechanistic insights into the interpretability of adversarial vulnerabilities. Inspired by these findings, we propose a neuron-level defense strategy, Targeted Neuron Tuning (TNT), which selectively finetunes the identified targeted neurons for specific attacks. Experimental evaluations across multiple LLM architectures and scales demonstrate that TNT substantially improves model robustness against a wide range of jailbreak attacks, achieving safe rates exceeding 90% and even approaching 100%, while preserving general task performance, enabling precise and robust safety interventions. Warning: This paper contains example data that may be harmful.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
Builds on30
- Direct Preference Optimization: Your Language Model is Secretly a Reward ModelRafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning et al.NeurIPS 2023 · 10,924 citations
- Measuring Massive Multitask Language UnderstandingDan Hendrycks, Collin Burns, Steven Basart, Andy Zou et al.ICLR 2021 · 7,905 citations
- Locating and Editing Factual Associations in GPTKevin Meng, David Bau, Alex Andonian, Yonatan BelinkovNeurIPS 2022 · 3,415 citations
- TruthfulQA: Measuring How Models Mimic Human FalsehoodsStephanie Lin, Jacob Hilton, Owain EvansACL 2022 · 3,228 citations
- WinoGrande: An Adversarial Winograd Schema Challenge at ScaleKeisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula, Yejin ChoiAAAI 2020 · 3,037 citations
Related papers
- SafeSeek: Universal Attribution of Safety Circuits in Language ModelsMiao Yu, Siyuan Fu, Moayad Aloqaily, Zhenhong Zhou et al.ICML 2026 · 3 citations
- From "Sure" to "Sorry": Detecting Jailbreak in Large Vision Language Model via JailNeuronsYuyou Gan, Qingming Li, Junhao Li, Zhi Chen et al.ICLR 2026
- Safety Alignment via Constrained Knowledge UnlearningZesheng Shi, Yucheng Zhou, Jing Li, Yuxin Jin et al.ACL 2025 · 8 citations
- Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank ModificationsBoyi Wei, Kaixuan Huang, Yangsibo Huang, Tinghao Xie et al.ICML 2024 · 215 citations
- Jailbreaking the Matrix: Nullspace Steering for Controlled Model SubversionVishal Pramanik, Maisha Maliha, Susmit Jha, Sumit Kumar JhaICLR 2026 · 3 citations
