Adversarial Attack Robustness on BeaverTails
0Attack Success RateBase+ART
Evaluation Results
| Method | Links | |
|---|---|---|
| Base+ARTModel=Gemma3-1B-IT, Attack Scenario=Baseline2026.05 | 0 | |
| SEAM+ARTModel=Gemma3-1B-IT, Attack Scenario=Baseline2026.05 | 0 | |
| TAR+ARTModel=Gemma3-1B-IT, Attack Scenario=Baseline2026.05 | 0 | |
| TAR+ARTModel=Gemma3-1B-IT, Attack Scenario=Prefill2026.05 | 1 | |
| SEAM+ARTModel=Gemma3-1B-IT, Attack Scenario=Prefill2026.05 | 2 | |
| TARModel=Gemma3-1B-IT, Attack Scenario=Prefill2026.05 | 4 | |
| Base+ARTModel=Gemma3-1B-IT, Attack Scenario=Prefill2026.05 | 5 | |
| TARModel=Gemma3-1B-IT, Attack Scenario=Baseline2026.05 | 5 | |
| BaseModel=Gemma3-1B-IT, Attack Scenario=Baseline2026.05 | 6 | |
| SEAMModel=Gemma3-1B-IT, Attack Scenario=Baseline2026.05 | 7 | |
| SEAMModel=Gemma3-1B-IT, Attack Scenario=Prefill2026.05 | 16 | |
| BaseModel=Gemma3-1B-IT, Attack Scenario=Prefill2026.05 | 20 | |
| Base+ARTModel=Gemma3-1B-IT, Attack Scenario=Abliterate2026.05 | 49 | |
| TAR+ARTModel=Gemma3-1B-IT, Attack Scenario=Abliterate2026.05 | 55 | |
| TAR+ARTModel=Gemma3-1B-IT, Attack Scenario=Abliterate + Prefill2026.05 | 56 | |
| SEAM+ARTModel=Gemma3-1B-IT, Attack Scenario=Abliterate2026.05 | 64 | |
| TARModel=Gemma3-1B-IT, Attack Scenario=Abliterate + Prefill2026.05 | 66 | |
| Base+ARTModel=Gemma3-1B-IT, Attack Scenario=Abliterate + Prefill2026.05 | 67 | |
| SEAM+ARTModel=Gemma3-1B-IT, Attack Scenario=Abliterate + Prefill2026.05 | 68 | |
| SEAMModel=Gemma3-1B-IT, Attack Scenario=Abliterate + Prefill2026.05 | 76 | |
| TARModel=Gemma3-1B-IT, Attack Scenario=Abliterate2026.05 | 76 | |
| SEAMModel=Gemma3-1B-IT, Attack Scenario=Abliterate2026.05 | 80 | |
| BaseModel=Gemma3-1B-IT, Attack Scenario=Abliterate2026.05 | 93 | |
| BaseModel=Gemma3-1B-IT, Attack Scenario=Abliterate + Prefill2026.05 | 96 |