AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

MESA: Improving MoE Safety Alignment via Decentralized Expertise

Yitong Sun, Yao Huang, Teng Li, Ranjie Duan, Yichi Zhang, Xingjun Ma, Hui Xue, Xingxing Wei

ICML 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
fine-tune all experts instead of OT-selected experts (ablation)MESA · DeepSeek-V2-Lite Math500accuracy: 24.80 -> 10.80Hurt
tune only the router, without expert updates (ablation)MESA · DeepSeek-V2-Lite WildJailbreaksafety rate: 60.20 vs 90.90 for full MESAHurt
add dynamic routing refinement to OT-selected experts (ablation)MESA · DeepSeek-V2-Lite GSM8Kaccuracy: 51.48 -> 61.00Improved
use full MESA routing refinement instead of plain router tuning with OT experts (ablation)MESA · Qwen3-30B-A3B Math500accuracy: 52.00 -> 91.00Improved
select experts by OT instead of middle-rank heuristic (ablation)MESA · Qwen3-30B-A3B Math500accuracy: 41.60 -> 52.00Improved
add shift parameter to cost function (alpha_shift 0 to 20) (ablation)MESA on DeepSeek · DeepSeek-V2-Lite Stratasafety rate: 92.50 -> 96.00Improved
raise alpha_shift too high (10 to 30) (ablation)MESA on Qwen · Qwen3-30B-A3B Stratasafety rate: 99.50 -> 89.50Hurt
use MESA instead of SFT or Stair-DPO on DeepSeeksafety alignment · DeepSeek-V2-Lite GSM8Kaccuracy: 16.15 -> 66.11Improved
use GRPO instead of SFT for safety tuningDeepSeek-V2-Lite · Stratasafety rate: 92.00 -> 64.00Hurt
MESA vs Stair-DPO on Qwen WildJailbreaksafety alignment · Qwen3-30B-A3B WildJailbreaksafety rate: 98.60 -> 97.65Hurt
use MESA instead of SafeX against routing attack F-SOURMoE safety · JailbreakBench on DeepSeekattack success rate: 15.38% -> 0.00%Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy