Kill Switch Evaluation on KILLBENCH System-Prompt Bypass Agent 1.0 (test)
31.07KSRAUTOGUARD
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AUTOGUARDModel=Grok-4.3, neff=5292025.09 | 31.07 | 32.32 | 58.16 | 6.91 | 54.61 | 3.8 | |
| IPIModel=GPT-5.2, neff=3712025.09 | 18.55 | 27.49 | 46.91 | 2 | 26.55 | 32.55 | |
| AUTOGUARDModel=GPT-5.2, neff=3732025.09 | 16.36 | 24.13 | 47.45 | 4 | 52.76 | 32.18 | |
| AUTOGUARDModel=Qwen3.6-35B-A3B, neff=5202025.09 | 12.75 | 13.48 | 75.62 | 6.18 | 11.47 | 5.45 | |
| WarningModel=GPT-5.2, neff=3852025.09 | 12.53 | 17.89 | 51.47 | 6.02 | 51.72 | 30 | |
| IPIModel=Grok-4.3, neff=5422025.09 | 6.16 | 6.25 | 83.45 | 8.93 | 8.35 | 1.45 | |
| WarningModel=Qwen3.6-35B-A3B, neff=5262025.09 | 4.73 | 4.94 | 87.27 | 3.65 | 31.87 | 4.36 | |
| GCGModel=GPT-5.2, neff=3382025.09 | 3.82 | 6.21 | 54.91 | 2.73 | 0.55 | 38.55 | |
| IPIModel=Qwen3.6-35B-A3B, neff=5262025.09 | 3.8 | 3.97 | 86.35 | 5.45 | 22.34 | 4.36 | |
| GCGModel=Grok-4.3, neff=5062025.09 | 2.89 | 3.14 | 87.07 | 2.02 | 0.16 | 8.02 | |
| WarningModel=Grok-4.3, neff=5212025.09 | 2.36 | 2.49 | 84 | 8.38 | 35.07 | 5.27 | |
| GCGModel=Qwen3.6-35B-A3B, neff=5212025.09 | 1.64 | 1.73 | 89.84 | 3.27 | 0 | 5.25 | |
| AUTOGUARDModel=Gemma4-31B-it, neff=5342025.09 | 1.25 | 1.29 | 95.29 | 0.55 | 30.32 | 0.73 | |
| WarningModel=Gemma4-31B-it, neff=5492025.09 | 0.55 | 0.55 | 95.29 | 3.98 | 5.05 | 0.18 | |
| IPIModel=Gemma4-31B-it, neff=5482025.09 | 0.39 | 0.39 | 96.48 | 2.8 | 64.93 | 0.27 | |
| GCGModel=Gemma4-31B-it, neff=5502025.09 | 0.18 | 0.18 | 99.09 | 0.73 | 0 | 0 |