AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

MAGIC: A Co-Evolving Attacker–Defender Adversarial Game for Robust LLM Safety

Xiaoyu Wen, Zhida He, Han Qi, Ziyu Wan, Zhongtian Ma, Ying Wen, Tianhang Zheng, Xingcheng Xu, Chaochao Lu, Qiaosheng Zhang

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Adversarial prompts for language models · 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 MAGIC adversarial co-evolution trainingQwen2.5-7B-Instruct · WildGuardTest adv harmASR: 0.365 -> 0.023Improved
add MAGIC instead of Self-RedTeamQwen2.5-7B-Instruct · WildGuardTest adv harmASR: 0.255 -> 0.023Improved
add MAGIC defender trainingQwen2.5-7B-Instruct · HarmBench GCG attackASR: 43.90 -> 11.25Improved
add MAGIC defender trainingQwen2.5-7B-Instruct · X-Teaming multi-turnMulti-turn ASR: 94.9% -> 85.4%Improved
add MAGIC defender trainingQwen2.5-7B-Instruct · AlpacaEval 2LC win rate: 33.733% -> 33.224%No clear effect
add MAGIC defender trainingQwen2.5-14B-Instruct · AlpacaEval 2LC win rate: 38.113% -> 30.954%Hurt
add the adversarial game (attacker) to defender training (ablation)Qwen2.5-7B-Instruct · WildJailBreak adv benigncompliance: 0.496 -> 0.968Improved
add the adversarial game (attacker) to defender training (ablation)Qwen2.5-7B-Instruct · WildJailBreak adv harmASR: 0.127 -> 0.198Hurt
add attacker training (unfreeze attacker) (ablation)Qwen2.5-7B-Instruct · AlpacaEval 2LC win rate: 23.491% -> 33.224%Improved
add attacker training (unfreeze attacker) (ablation)Qwen2.5-7B-Instruct · XSTest all-safecompliance: 0.812 -> 0.945Improved
add trained reasoning attacker instead of base model attacker (ablation)Qwen2.5-7B-Instruct · WildJailBreak adv benigncompliance: 0.876 -> 0.968Improved
add trained reasoning attacker instead of base model attacker (ablation)Qwen2.5-7B-Instruct · WildJailBreak adv harmASR: 0.080 -> 0.198Hurt

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