Jailbreaking LLMs on HarmBench target: Mistral-7B-v2-RR
70.31Attack Success RateTHREAT (Challenge)
Evaluation results
| Method | Links | |
|---|---|---|
| THREAT (Challenge)Meaningful=true2026.05 | 70.31 | |
| Adversarial ReasoningMeaningful=true2026.05 | 70 | |
| THREAT (Address)Meaningful=true2026.05 | 54.06 | |
| TAP-TMeaningful=true2026.05 | 40 | |
| PAIRMeaningful=true2026.05 | 32 | |
| GCGMeaningful=false2026.05 | 6 | |
| Prompt + Random SearchMeaningful=false2026.05 | 0 |