AI Paper Atlas

Research map / Large language models / Alignment and preferences

Towards Context-Invariant Safety Alignment for Large Language Models

Yixu Wang, Yang Yao, Xin Wang, Yifeng Gao, Yan Teng, Xingjun Ma, Yingchun Wang

ICML 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior · 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
add AIR to GRPOGRPO, Qwen-2.5-14B · Average OOD (3 domains)group accuracy: 23.05% -> 56.54%Improved
add AIR to GSPOGSPO, Qwen-2.5-14B · Average OOD (3 domains)group accuracy: 30.22% -> 54.77%Improved
add AIR to GRPOGRPO, Qwen-2.5-14B · Safety OODgroup accuracy: 13.73% -> 60.78%Improved
add AIR to GRPOGRPO, Qwen-2.5-14B · Moral OODgroup accuracy: 14.71% -> 47.79%Improved
add symmetric V-REx penalty to GRPO (ablation)GRPO, Qwen-2.5-14B · Safety IDgroup accuracy: 71.15% -> 35.40%Hurt
add symmetric V-REx penalty to GRPO (ablation)GRPO, Qwen-2.5-14B · Moral IDgroup accuracy: 34.31% -> 7.15%Hurt
add symmetric V-REx penalty to GRPO (ablation)GRPO, Qwen-2.5-14B · Math IDgroup accuracy: 64.60% -> 60.71%Mixed
add AIR to GRPOGRPO, Qwen-2.5-14B · Math IDgroup accuracy: 64.60% -> 63.72%No clear effect
add AIR to GSPOGSPO, Qwen-2.5-14B · Math OODgroup accuracy: 54.87% -> 51.33%Hurt
use AIR coefficient lambda moderate (about 8e-4 to 1e-3) instead of 0 or very large (ablation)GRPO+AIR · Mixed ID and OOD averageAvg Acc and Avg group accuracyMixed
add AIR to GRPOGRPO, Qwen-2.5-14B · Latent dispersionL2 distance to intent centroid: 86.47 -> 71.54Improved
add AIR when open-ended reward is a gameable regex proxyGRPO, Qwen-2.5-14B · Safety, regex-reward stress testoracle judge safetyImproved

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