Safety Evaluation on Jailbreak Bench
3ASRChat model
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Chat modelBackbone LLM=LLaMa-2 7B2026.01 | 3 | 37 | |
| SafeReActModel=Finance-Llama3.1-8B-Instruct2026.03 | 3 | — | |
| Chat modelBackbone LLM=Qwen-2.5 7B2026.01 | 10 | 62 | |
| ITIBackbone LLM=Qwen-2.5 7B2026.01 | 10 | 56 | |
| Chat modelBackbone LLM=LLaMa-3.1 8B2026.01 | 13 | 77.7 | |
| ARREST-SCBackbone LLM=LLaMa-3.1 8B2026.01 | 17 | 85 | |
| ARREST-SCBackbone LLM=Qwen-2.5 7B2026.01 | 22 | 80 | |
| ITIBackbone LLM=LLaMa-3.1 8B2026.01 | 26 | 70 | |
| ARREST-SBBackbone LLM=LLaMa-3.1 8B2026.01 | 29 | 79 | |
| Chat modelBackbone LLM=Yi-1.5 9B2026.01 | 30 | 70 | |
| ARREST-SBBackbone LLM=Qwen-2.5 7B2026.01 | 36 | 79 | |
| ARREST-SCBackbone LLM=LLaMa-2 7B2026.01 | 42 | 94 | |
| Base ModelBackbone LLM=LLaMa-3.1 8B2026.01 | 42 | 70 | |
| Base ModelBackbone LLM=Qwen-2.5 7B2026.01 | 42 | 55 | |
| OriginalModel=Finance-Llama3.1-8B-Instruct2026.03 | 47 | — | |
| ARREST-SCBackbone LLM=Yi-1.5 9B2026.01 | 52 | 75 | |
| ARREST-SBBackbone LLM=LLaMa-2 7B2026.01 | 54 | 89 | |
| ARREST-SBBackbone LLM=Yi-1.5 9B2026.01 | 63 | 80 | |
| ITIBackbone LLM=Yi-1.5 9B2026.01 | 64 | 70 | |
| ITIBackbone LLM=LLaMa-2 7B2026.01 | 75 | 77 | |
| Base ModelBackbone LLM=Yi-1.5 9B2026.01 | 77 | 35 | |
| Base ModelBackbone LLM=LLaMa-2 7B2026.01 | 83 | 31 |