AI Paper Atlas

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

CAGE: A Framework for Culturally Adaptive Red-Teaming Benchmark Generation

Chaeyun Kim, YongTaek Lim, Kihyun Kim, Junghwan Kim, Minwoo Kim

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 2 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
generate Korean prompts with CAGE instead of direct translationLlama-3.1-8B Direct Request · KorSETASR (%): 28.2 -> 43.8Improved
generate Korean prompts with CAGE instead of direct translationQwen2.5-7B Direct Request · KorSETASR (%): 14.6 -> 25.3Improved
CAGE instead of LLM-Adaptation or template fillingKorean prompts · KorSET ASR across modelsASR (%): e.g. gemma2 Direct 9.8/9.3 -> 20.1Improved
CAGE instead of direct translationgemma2 AutoDAN · KorSETASR (%): 16.7 -> 35.4Improved
add specificity in English context (ablation)Llama-3.1-8B · 2x2 decomposition, Direct RequestASR (%): 16.68 -> 8.62Hurt
add specificity in Korean context (ablation)Llama-3.1-8B · 2x2 decomposition, Direct RequestASR (%): 32.36 -> 43.77Improved
shift from English to Korean cultural context at same specificity (ablation)Llama-3.1-8B · 2x2 decomposition, Direct RequestASR (%): 8.62 -> 43.77 (+35.16)Improved
shift from English to Korean cultural context (ablation)EXAONE-3.5-7.8B · 2x2 decomposition, Direct RequestASR (%): 21.88 -> 23.12 (+1.24)No clear effect
apply CAGE to Khmer instead of direct translationgemma3-12B-it · Khmer, category L Direct RequestASR (%): 2.7 -> 35.1Improved
apply CAGE to Khmer for prompt qualityLLM judge · Khmer category Aquality score: 3.55 -> 6.43Improved
increase target model sizeOpen-source LLM families 2B-32B · KorSETASRNo clear effect
transfer AutoDAN instead of GCG to black-box modelsGPT-4o via Llama3.1-8B · KorSET transferASR: 0.1473 -> 0.1592Improved

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