Single-turn Jailbreak Attack on HarmBench
91ASRVernacular
Evaluation Results
| Method | Links | |
|---|---|---|
| VernacularModel=Qwen3.5-9B, Variant=Mean2026.06 | 91 | |
| VernacularModel=Qwen3.5-4B, Variant=Mean2026.06 | 91 | |
| MimicryNarrativeModel=Qwen3.5-9B, Variant=PlanX2026.06 | 89 | |
| VernacularModel=Llama-3.1-8B, Variant=Mean2026.06 | 84 | |
| VernacularModel=AVG, Variant=Mean2026.06 | 74 | |
| VernacularModel=Gemma-4-E2B, Variant=Mean2026.06 | 73 | |
| VernacularModel=Gemma-4-E4B, Variant=Mean2026.06 | 68 | |
| VernacularModel=Llama-3.2-3B, Variant=Mean2026.06 | 67 | |
| VernacularModel=Gemini-2.5-Pro, Variant=Mean2026.06 | 67 | |
| MimicryNarrativeModel=AVG, Variant=PlanX2026.06 | 66 | |
| RolebreakPersonaModel=DeepSeek-V4-Pro, Variant=base2026.06 | 60 | |
| MimicryNarrativeModel=Qwen3.5-9B, Variant=base2026.06 | 57 | |
| VernacularModel=DeepSeek-V4-Pro, Variant=Mean2026.06 | 53 | |
| AdvPoetryModel=AVG, Variant=PlanX2026.06 | 46 | |
| MimicryNarrativeModel=AVG, Variant=base2026.06 | 36 | |
| AdvPoetryModel=AVG, Variant=base2026.06 | 36 | |
| CrescendoCondensedModel=AVG, Variant=PlanX2026.06 | 31 | |
| DeepInceptionModel=AVG, Variant=PlanX2026.06 | 31 | |
| RolebreakPersonaModel=Llama-3.1-8B, Variant=base2026.06 | 26 | |
| DeepInceptionModel=AVG, Variant=base2026.06 | 26 | |
| DaggerHappyEndingModel=AVG, Variant=base2026.06 | 26 | |
| DaggerHappyEndingModel=AVG, Variant=PlanX2026.06 | 25 | |
| RolebreakPersonaModel=Gemini-2.5-Pro, Variant=PlanX2026.06 | 20 | |
| RolebreakPersonaModel=DeepSeek-V4-Pro, Variant=PlanX2026.06 | 20 | |
| RolebreakPersonaModel=Llama-3.1-8B, Variant=PlanX2026.06 | 17 | |
| CrescendoCondensedModel=AVG, Variant=base2026.06 | 17 | |
| RolebreakPersonaModel=AVG, Variant=base2026.06 | 14 | |
| RolebreakPersonaModel=Gemini-2.5-Pro, Variant=base2026.06 | 10 | |
| RolebreakPersonaModel=AVG, Variant=PlanX2026.06 | 8 | |
| RolebreakPersonaModel=Gemma-4-E4B, Variant=base2026.06 | 7 | |
| CrescendoCondensedModel=Qwen3.5-9B, Variant=PlanX2026.06 | 7 | |
| RolebreakPersonaModel=Gemma-4-E4B, Variant=PlanX2026.06 | 6 | |
| RolebreakPersonaModel=Gemma-4-E2B, Variant=base2026.06 | 4 | |
| CrescendoCondensedModel=Qwen3.5-9B, Variant=base2026.06 | 4 | |
| RolebreakPersonaModel=Llama-3.2-3B, Variant=base2026.06 | 2 | |
| RolebreakPersonaModel=Gemma-4-E2B, Variant=PlanX2026.06 | 2 | |
| RolebreakPersonaModel=Qwen3.5-4B, Variant=PlanX2026.06 | 1 | |
| RolebreakPersonaModel=Llama-3.2-3B, Variant=PlanX2026.06 | 1 | |
| RolebreakPersonaModel=Qwen3.5-9B, Variant=base2026.06 | 0 | |
| RolebreakPersonaModel=Qwen3.5-9B, Variant=PlanX2026.06 | 0 | |
| RolebreakPersonaModel=Qwen3.5-4B, Variant=base2026.06 | 0 |