Jailbreaking LLMs on HarmBench target: LLaMA-2-7B
72.81Attack Success Rate (ASR)THREAT (Address)
Evaluation results
| Method | Links | |
|---|---|---|
| THREAT (Address)Meaningful=true2026.05 | 72.81 | |
| THREAT (Challenge)Meaningful=true2026.05 | 72.5 | |
| Adversarial ReasoningMeaningful=true2026.05 | 60 | |
| Prompt + Random SearchMeaningful=false2026.05 | 48 | |
| TAP-TMeaningful=true2026.05 | 48 | |
| AutoDAN-TurboMeaningful=true2026.05 | 36 | |
| PAIRMeaningful=true2026.05 | 34 | |
| GCGMeaningful=false2026.05 | 32 |