AI Paper Atlas

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

Sparse Models, Sparse Safety: Unsafe Routes in Mixture-of-Experts LLMs

Yukun Jiang, Hai Huang, Mingjie Li, Yage Zhang, Michael Backes, Yang Zhang

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
manipulate 1 sample-level high-RoSais routerRoSais attack (avg of 4 MoE LLMs) · JailbreakBenchASR: 0.16 -> 0.49Improved
raise changed layers from 1 to 5 (sample level) (ablation)RoSais attack on DeepSeek-V2-Lite · JailbreakBenchASR: 0.50 -> 0.46Hurt
raise changed layers from 1 to 5 (sample level) (ablation)RoSais attack on Mixtral-8x7B · JailbreakBenchASR: 0.66 -> 0.81Improved
raise changed layers from 1 to 5 (dataset level) (ablation)RoSais universal route on DeepSeek-V2-Lite · JailbreakBenchASR: 0.27 -> 0.79Improved
use RoSais-guided routers instead of random layers and masks (ablation)RoSais attack on DeepSeek-V2-Lite · JailbreakBenchASR: 0.24 -> 0.79Improved
use RoSais-guided routers instead of random layers and masks (ablation)RoSais attack on DeepSeek-V2-Lite · AdvBenchASR: 0.14 -> 0.90Improved
replace RoSais top-5 with F-SOUR token-layer-wise searchRoSais-based attack (avg of 4 MoE LLMs) · JailbreakBenchASR: 0.59 -> 0.90Improved
increase F-SOUR restarts S4 from 1 to 5 (ablation)F-SOUR on DeepSeek-V2-Lite · JailbreakBenchASR: 0.85 -> 0.94Improved
increase F-SOUR restarts S4 from 1 to 5 (ablation)F-SOUR on DeepSeek-V2-Lite · AdvBenchASR: 0.90 -> 1.00Improved
increase randomizations per token-layer pair S3 from 5 to 10 (ablation)F-SOUR on DeepSeek-V2-Lite · JailbreakBenchASR: 0.92 -> 0.94No clear effect
add PAT defenseDataset-level RoSais attack on DeepSeek-V2-Lite · JailbreakBenchASR: 0.79 -> 0.16Hurt
add PAT defenseF-SOUR on DeepSeek-V2-Lite · JailbreakBenchASR: 0.94 -> 0.90No clear effect

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