@inproceedings{wang-etal-2026-chain-mesh,
title = "From Chain to Mesh: Evolving Robust Safety Alignment via Neuron-Circuit Topology",
author = "Wang, Guan and
Tang, Xuehai and
Zhou, Biyu and
Xiao, Wenjie and
Zhao, Yuanshu and
Lv, Lijia and
Han, Jizhong and
Hu, Songlin",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2555/",
pages = "58079--58100",
ISBN = "979-8-89176-459-0",
abstract = "Despite substantial progress in safety alignment, aligned Large Language Models (LLMs) remain fragile under small parameter-space and activation-space perturbations, such as downstream fine-tuning and jailbreak attacks. To explain why local perturbations can trigger broad safety failures, we analyze safety-task neuron circuits from the perspectives of activation flow and circuit topology. We identify a critical topological fragility in safety alignment: In aligned models, safety-task neuron circuits exhibit a low-redundancy, isolated single-chain topology, in sharp contrast to the highly redundant dense-mesh topology of general-task circuits. This brittle topology creates rigid serial dependencies, whereby perturbing a small set of upstream neurons can disrupt downstream safety neurons and cascade into broad alignment failures. To mitigate this topological fragility, we propose Dense Safety Mesh (DSM), a topology-aware alignment framework that transforms brittle single-chain safety circuits into more redundant and resilient mesh-like structures. DSM employs structural adversarial masking to suppress dominant safety pathways, forcing safety circuits to develop diverse parallel pathways. Extensive experiments across seven models and multiple evaluations show that DSM improves robustness against perturbations attacks while maintaining general capabilities."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-chain-mesh">
<titleInfo>
<title>From Chain to Mesh: Evolving Robust Safety Alignment via Neuron-Circuit Topology</title>
</titleInfo>
<name type="personal">
<namePart type="given">Guan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuehai</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Biyu</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenjie</namePart>
<namePart type="family">Xiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuanshu</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lijia</namePart>
<namePart type="family">Lv</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jizhong</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Songlin</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Despite substantial progress in safety alignment, aligned Large Language Models (LLMs) remain fragile under small parameter-space and activation-space perturbations, such as downstream fine-tuning and jailbreak attacks. To explain why local perturbations can trigger broad safety failures, we analyze safety-task neuron circuits from the perspectives of activation flow and circuit topology. We identify a critical topological fragility in safety alignment: In aligned models, safety-task neuron circuits exhibit a low-redundancy, isolated single-chain topology, in sharp contrast to the highly redundant dense-mesh topology of general-task circuits. This brittle topology creates rigid serial dependencies, whereby perturbing a small set of upstream neurons can disrupt downstream safety neurons and cascade into broad alignment failures. To mitigate this topological fragility, we propose Dense Safety Mesh (DSM), a topology-aware alignment framework that transforms brittle single-chain safety circuits into more redundant and resilient mesh-like structures. DSM employs structural adversarial masking to suppress dominant safety pathways, forcing safety circuits to develop diverse parallel pathways. Extensive experiments across seven models and multiple evaluations show that DSM improves robustness against perturbations attacks while maintaining general capabilities.</abstract>
<identifier type="citekey">wang-etal-2026-chain-mesh</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2555/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>58079</start>
<end>58100</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T From Chain to Mesh: Evolving Robust Safety Alignment via Neuron-Circuit Topology
%A Wang, Guan
%A Tang, Xuehai
%A Zhou, Biyu
%A Xiao, Wenjie
%A Zhao, Yuanshu
%A Lv, Lijia
%A Han, Jizhong
%A Hu, Songlin
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F wang-etal-2026-chain-mesh
%X Despite substantial progress in safety alignment, aligned Large Language Models (LLMs) remain fragile under small parameter-space and activation-space perturbations, such as downstream fine-tuning and jailbreak attacks. To explain why local perturbations can trigger broad safety failures, we analyze safety-task neuron circuits from the perspectives of activation flow and circuit topology. We identify a critical topological fragility in safety alignment: In aligned models, safety-task neuron circuits exhibit a low-redundancy, isolated single-chain topology, in sharp contrast to the highly redundant dense-mesh topology of general-task circuits. This brittle topology creates rigid serial dependencies, whereby perturbing a small set of upstream neurons can disrupt downstream safety neurons and cascade into broad alignment failures. To mitigate this topological fragility, we propose Dense Safety Mesh (DSM), a topology-aware alignment framework that transforms brittle single-chain safety circuits into more redundant and resilient mesh-like structures. DSM employs structural adversarial masking to suppress dominant safety pathways, forcing safety circuits to develop diverse parallel pathways. Extensive experiments across seven models and multiple evaluations show that DSM improves robustness against perturbations attacks while maintaining general capabilities.
%U https://aclanthology.org/2026.emnlp-main.2555/
%P 58079-58100
Markdown (Informal)
[From Chain to Mesh: Evolving Robust Safety Alignment via Neuron-Circuit Topology](https://aclanthology.org/2026.emnlp-main.2555/) (Wang et al., EMNLP 2026)
ACL
- Guan Wang, Xuehai Tang, Biyu Zhou, Wenjie Xiao, Yuanshu Zhao, Lijia Lv, Jizhong Han, and Songlin Hu. 2026. From Chain to Mesh: Evolving Robust Safety Alignment via Neuron-Circuit Topology. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 58079–58100, Budapest, Hungary. Association for Computational Linguistics.