MESA: Improving MoE Safety Alignment via Decentralized Expertise

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Sun, Yitong, Huang, Yao, Li, Teng, Duan, Ranjie, Zhang, Yichi, Ma, Xingjun, Xue, Hui, Wei, Xingxing
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913175761321984
author Sun, Yitong
Huang, Yao
Li, Teng
Duan, Ranjie
Zhang, Yichi
Ma, Xingjun
Xue, Hui
Wei, Xingxing
author_facet Sun, Yitong
Huang, Yao
Li, Teng
Duan, Ranjie
Zhang, Yichi
Ma, Xingjun
Xue, Hui
Wei, Xingxing
contents Mixture-of-Experts (MoE) architectures scale Large Language Models (LLMs) efficiently, enabling greater capacity with reduced computational cost by dynamically routing inputs to relevant experts, yet introduce a critical vulnerability: Safety Sparsity, where safety capabilities concentrate in few experts, making them susceptible to adversarial bypassing. Meanwhile, conventional alignment methods uniformly adapt all parameters, ignoring their functional differences and inadvertently degrading performances. To address these challenges, we propose MESA (MoE Safety Alignment), a targeted alignment framework for MoE-based LLMs that strategically decentralizes safety responsibility to maximize coverage while minimizing interference with utility. Based on Optimal Transport (OT) theory, MESA operates through two mechanisms: (1) Expert Capacity Reallocation uses a transport cost matrix to distribute safety duties to the most cost-effective experts, and (2) Dynamic Routing Refinement constrains the router to precisely activate these decentralized modules. Experiments show that MESA achieves robust defensive performance against varied harmful benchmarks while preserving helpfulness. Code is available at https://github.com/lorraine021/MESA.
format Preprint
id arxiv_https___arxiv_org_abs_2606_00651
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MESA: Improving MoE Safety Alignment via Decentralized Expertise
Sun, Yitong
Huang, Yao
Li, Teng
Duan, Ranjie
Zhang, Yichi
Ma, Xingjun
Xue, Hui
Wei, Xingxing
Machine Learning
Artificial Intelligence
Computation and Language
Mixture-of-Experts (MoE) architectures scale Large Language Models (LLMs) efficiently, enabling greater capacity with reduced computational cost by dynamically routing inputs to relevant experts, yet introduce a critical vulnerability: Safety Sparsity, where safety capabilities concentrate in few experts, making them susceptible to adversarial bypassing. Meanwhile, conventional alignment methods uniformly adapt all parameters, ignoring their functional differences and inadvertently degrading performances. To address these challenges, we propose MESA (MoE Safety Alignment), a targeted alignment framework for MoE-based LLMs that strategically decentralizes safety responsibility to maximize coverage while minimizing interference with utility. Based on Optimal Transport (OT) theory, MESA operates through two mechanisms: (1) Expert Capacity Reallocation uses a transport cost matrix to distribute safety duties to the most cost-effective experts, and (2) Dynamic Routing Refinement constrains the router to precisely activate these decentralized modules. Experiments show that MESA achieves robust defensive performance against varied harmful benchmarks while preserving helpfulness. Code is available at https://github.com/lorraine021/MESA.
title MESA: Improving MoE Safety Alignment via Decentralized Expertise
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2606.00651