LLM Jailbreaking on HarmBench target: LLaMA-3-8B-RR
67.81Attack Success RateTHREAT (Challenge)
Evaluation Results
| Method | Links | |
|---|---|---|
| THREAT (Challenge)Meaningful=true2026.05 | 67.81 | |
| THREAT (Address)Meaningful=true2026.05 | 64.38 | |
| Adversarial ReasoningMeaningful=true2026.05 | 44 | |
| TAP-TMeaningful=true2026.05 | 32 | |
| PAIRMeaningful=true2026.05 | 22 | |
| GCGMeaningful=false2026.05 | 2 | |
| Prompt + Random SearchMeaningful=false2026.05 | 0 |