Jailbreaking on JailbreakBench (Model-Specific ASR Evaluation)
94.55ASR@1 (Qwen2.5-7B-IT)TRACE (single)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TRACE (single)Workflow Category=Multi-turn RL, Judge Model=GPT-4o2026.05 | 94.55 | 75.75 | 72.12 | |
| TRACE (mix)Workflow Category=Multi-turn RL, Judge Model=GPT-4o2026.05 | 85.84 | 69.69 | 75.15 | |
| SirenWorkflow Category=Multi-turn RL, Judge Model=GPT-4o2026.05 | 81.82 | 56.36 | 65.54 | |
| TROJailWorkflow Category=Multi-turn RL, Judge Model=GPT-4o2026.05 | 78.18 | 35.76 | 74.55 | |
| CrescendoWorkflow Category=Multi-turn Workflow, Judge Model=GPT-4o2026.05 | 65.45 | 7.27 | 5.46 | |
| AutoDan-TurboWorkflow Category=Single-turn, Judge Model=GPT-4o2026.05 | 56.36 | 23.63 | 5.45 | |
| X-TeamingWorkflow Category=Multi-turn Workflow, Judge Model=GPT-4o2026.05 | 52.72 | 34.54 | 30.91 | |
| Jailbreak-R1Workflow Category=Single-turn, Judge Model=GPT-4o2026.05 | 49.09 | 36.36 | 7.27 | |
| MUSE-AWorkflow Category=Multi-turn Workflow, Judge Model=GPT-4o2026.05 | 41.81 | 13.73 | 3.64 | |
| ActorAttackWorkflow Category=Multi-turn Workflow, Judge Model=GPT-4o2026.05 | 40 | 30.91 | 34.55 | |
| PAIRWorkflow Category=Single-turn, Judge Model=GPT-4o2026.05 | 26.36 | 21.82 | 1.82 |