Black-Box LLM Jailbreaking on HarmBench (test)
98.74ASR (%)TrailBlazer
Evaluation Results
| Method | Links | |
|---|---|---|
| TrailBlazerTarget LLM=Qwen 3-14B2026.02 | 98.74 | |
| TrailBlazerTarget LLM=Gemma 3-12B2026.02 | 98.11 | |
| RLbreakerTarget LLM=Gemma 3-12B2026.02 | 94.96 | |
| TrailBlazerTarget LLM=Llama 3.2-11B2026.02 | 91.19 | |
| RLbreakerTarget LLM=Qwen 3-14B2026.02 | 73.58 | |
| TrailBlazerTarget LLM=GPT-oss-20B2026.02 | 54.09 | |
| RLbreakerTarget LLM=Llama 3.2-11B2026.02 | 50.31 | |
| RLbreakerTarget LLM=GPT-oss-20B2026.02 | 5.03 |