Attack Success Rate Evaluation on SKILL-INJECT Contextual
0Attack Success Rate (ASR)Direct Attack
Evaluation Results
| Method | Links | |
|---|---|---|
| Direct AttackModel=kimi-k2.52026.04 | 0 | |
| Direct AttackModel=claude-sonnet-4-5-202509292026.04 | 2 | |
| Direct AttackModel=Qwen3.5-Plus2026.04 | 2 | |
| Direct AttackModel=glm-52026.04 | 2 | |
| Direct AttackModel=MiniMax-M2.52026.04 | 4 | |
| Direct AttackModel=gpt-5.42026.04 | 5 | |
| SKILL-INJECTModel=kimi-k2.52026.04 | 5 | |
| SKILL-INJECTModel=Qwen3.5-Plus2026.04 | 5 | |
| Direct AttackModel=gemini-3.0-pro-preview2026.04 | 7 | |
| SKILL-INJECTModel=MiniMax-M2.52026.04 | 7 | |
| Direct AttackModel=gpt-5.4-nano2026.04 | 10 | |
| SKILL-INJECTModel=gemini-3.0-pro-preview2026.04 | 10 | |
| Direct AttackModel=doubao-seed-1-8-2512282026.04 | 10 | |
| SKILL-INJECTModel=gpt-5.4-nano2026.04 | 12 | |
| Direct AttackModel=hunyuan-2.0-instruct-202511112026.04 | 12 | |
| SKILL-INJECTModel=glm-52026.04 | 15 | |
| SKILL-INJECTModel=doubao-seed-1-8-2512282026.04 | 15 | |
| SKILL-INJECTModel=gpt-5.42026.04 | 19 | |
| SKILL-INJECTModel=claude-sonnet-4-5-202509292026.04 | 20 | |
| SKILL-INJECTModel=hunyuan-2.0-instruct-202511112026.04 | 20 | |
| SkillAttackModel=Qwen3.5-Plus2026.04 | 56 | |
| SkillAttackModel=claude-sonnet-4-5-202509292026.04 | 64 | |
| SkillAttackModel=gemini-3.0-pro-preview2026.04 | 68 | |
| SkillAttackModel=glm-52026.04 | 68 | |
| SkillAttackModel=gpt-5.42026.04 | 71 | |
| SkillAttackModel=doubao-seed-1-8-2512282026.04 | 71 | |
| SkillAttackModel=gpt-5.4-nano2026.04 | 76 | |
| SkillAttackModel=MiniMax-M2.52026.04 | 78 | |
| SkillAttackModel=kimi-k2.52026.04 | 85 | |
| SkillAttackModel=hunyuan-2.0-instruct-202511112026.04 | 88 |