Jailbreak Attack on Harmful behaviors jailbreak evaluation set
100ASR (Harmful Behaviors)Baseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BaselineTarget Model=Qwen3-8B2026.05 | 100 | |
| BaselineTarget Model=Qwen3-1.7B2026.05 | 100 | |
| BaselineTarget Model=Phi-4-reasoning2026.05 | 97 | |
| BaselineTarget Model=GPT-OSS-20B2026.05 | 94 | |
| BaselineTarget Model=Gemma-4-E4B-it2026.05 | 88 | |
| BCTTarget Model=Gemma-4-E4B-it2026.05 | 57 | |
| BCTTarget Model=Qwen3-1.7B2026.05 | 36 | |
| ACTTarget Model=Gemma-4-E4B-it2026.05 | 33 | |
| BCTTarget Model=Qwen3-8B2026.05 | 30 | |
| BCTTarget Model=Phi-4-reasoning2026.05 | 15.3 | |
| ACTTarget Model=Qwen3-1.7B2026.05 | 14 | |
| ACTTarget Model=GPT-OSS-20B2026.05 | 1 | |
| BCTTarget Model=GPT-OSS-20B2026.05 | 1 | |
| ACTTarget Model=Qwen3-8B2026.05 | 0 | |
| ACTTarget Model=Phi-4-reasoning2026.05 | 0 |