AI Paper Atlas

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

Multilingual Safety Alignment via Representation-Space Separability

Dan Shi, Zhuowen Han, Deyi Xiong

ICML 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs

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
use SMO instead of SimPO for multilingual safety alignmentLLaMA-3.1-8B-Instruct · MultiJail avg ASRattack success rate (%): 5.87 -> 1.53Improved
add SMG example-wise modulation of the SimPO loss (ablation)SMO on LLaMA-3.1-8B · MultiJail avg ASRattack success rate (%): 3.49 -> 1.53Improved
add SMG example-wise modulation of the SimPO loss (ablation)SMO on LLaMA-3.1-8B · AdvBench-X avg ASRattack success rate (%): 2.53 -> 0.29Improved
add SMG example-wise modulation of the SimPO loss (ablation)SMO on Qwen-2.5-7B · MultiJail avg ASRattack success rate (%): 29.42 -> 21.22Improved
add SMG example-wise modulation of the SimPO loss (ablation)SMO on Qwen-2.5-7B · AdvBench-X avg ASRattack success rate (%): 16.41 -> 9.97Improved
add margin loss term (ablation)SMO on LLaMA-3.1-8B · MultiJail and AdvBench-X ASRattack success rate (%): 1.32 -> 1.53 (MultiJail)No clear effect
add margin loss term (ablation)SMO on LLaMA-3.1-8B · MGSM accuracyaccuracy (%): 24.47 -> 25.24Improved
use Chinese instead of English as anchor language (ablation)SMO on Qwen2.5-7B-Instruct · MultiJail avg ASRattack success rate (%): 21.22 -> 20.21No clear effect
train with SMOLLaMA-3.1-8B-Instruct · M-MMLU avg accuracyaccuracy (%): 51.63 -> 51.39No clear effect
train with SMOLLaMA-3.1-8B-Instruct · MGSM multilingual avgaccuracy (%): 26.49 -> 23.07Hurt
train with SMOQwen-2.5-7B-Instruct · MGSM Enaccuracy (%): 70.40 -> 73.20Improved
train with SMOLLaMA-3.1-8B-Instruct · harmful-harmless L2 distance on training data (avg)L2 distance: 23.39 -> 25.69Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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