Jailbreak Defense on HarmBench (test)
0ASRBCT
Evaluation Results
| Method | Links | |
|---|---|---|
| BCTBase Model=GPT-OSS-20B2026.05 | 0 | |
| ACTBase Model=Qwen3-8B2026.05 | 0 | |
| BCTBase Model=Phi-4-reasoning2026.05 | 0 | |
| ACTBase Model=Phi-4-reasoning2026.05 | 1 | |
| ACTBase Model=Qwen3-1.7B2026.05 | 2.4 | |
| ACTBase Model=GPT-OSS-20B2026.05 | 4.5 | |
| ACTBase Model=Gemma-4-E4B-it2026.05 | 11.2 | |
| BaselineBase Model=GPT-OSS-20B2026.05 | 12.2 | |
| BCTBase Model=Gemma-4-E4B-it2026.05 | 14.3 | |
| BCTBase Model=Qwen3-8B2026.05 | 18.4 | |
| BaselineBase Model=Phi-4-reasoning2026.05 | 30 | |
| BCTBase Model=Qwen3-1.7B2026.05 | 51 | |
| BaselineBase Model=Qwen3-8B2026.05 | 62.6 | |
| BaselineBase Model=Qwen3-1.7B2026.05 | 63.2 | |
| BaselineBase Model=Gemma-4-E4B-it2026.05 | 73.5 |