Safety evaluation on Sorry-Bench
99.09Safety ScoreIDGAF
Evaluation Results
| Method | Links | |
|---|---|---|
| IDGAFModel=Llama-2-13B2026.01 | 99.09 | |
| AOAModel=Llama-2-13B2026.01 | 98.41 | |
| SkeletonModel=Llama-2-13B2026.01 | 97.5 | |
| IDGAFModel=Gemma-1.1-7B2026.01 | 96.14 | |
| FormalModel=Qwen3-4B-think2026.01 | 93.86 | |
| IDGAFModel=Qwen3-4B-think2026.01 | 93.64 | |
| AOAModel=Qwen3-4B-think2026.01 | 93.18 | |
| SkeletonModel=Qwen3-4B-think2026.01 | 93.18 | |
| AOAModel=Llama-3.1-8B2026.01 | 92.27 | |
| BaseModel=Qwen3-4B-think2026.01 | 90.23 | |
| SkeletonModel=Gemma-1.1-7B2026.01 | 89.32 | |
| BaseModel=Gemma-2-9B2026.01 | 89.09 | |
| BaseModel=Llama-3.2-1B2026.01 | 88.86 | |
| AOAModel=Gemma-2-9B2026.01 | 88.64 | |
| BaseModel=Llama-2-13B2026.01 | 88.41 | |
| GCGModel=Qwen3-4B-think2026.01 | 88.38 | |
| AOAModel=Gemma-1.1-7B2026.01 | 87.73 | |
| SkeletonBackbone=GPT-oss-20B2026.01 | 87.73 | |
| FormalModel=Gemma-2-9B2026.01 | 87.05 | |
| SkeletonModel=Gemma-2-9B2026.01 | 86.36 | |
| Refusal SuppressionModel=Gemma-2-9B2026.01 | 86.36 | |
| IDGAFModel=Llama-3.1-8B2026.01 | 85.91 | |
| BaseBackbone=GPT-oss-20B2026.01 | 85.45 | |
| AOABackbone=GPT-oss-20B2026.01 | 85.45 | |
| FormalModel=Llama-2-13B2026.01 | 84.77 | |
| GCGModel=Llama-2-13B2026.01 | 84.55 | |
| IDGAFModel=Gemma-2-9B2026.01 | 84.09 | |
| BaseModel=Gemma-1.1-7B2026.01 | 83.18 | |
| GCGModel=Gemma-2-9B2026.01 | 83.18 | |
| IDGAFBackbone=GPT-oss-20B2026.01 | 82.73 | |
| AOAModel=Llama-3.2-1B2026.01 | 82.05 | |
| FormalModel=Gemma-1.1-7B2026.01 | 81.36 | |
| BaseModel=Gemma-2-2B2026.01 | 81.36 | |
| FormalBackbone=GPT-oss-20B2026.01 | 80.45 | |
| FormalModel=Llama-3.1-8B2026.01 | 79.09 | |
| BaseModel=Llama-3.1-8B2026.01 | 78.86 | |
| GCGBackbone=GPT-oss-20B2026.01 | 78.41 | |
| GCGModel=Llama-3.2-1B2026.01 | 77.5 | |
| FormalModel=Gemma-2-2B2026.01 | 77.27 | |
| GCGModel=Gemma-1.1-7B2026.01 | 76.14 | |
| Refusal SuppressionModel=Llama-2-13B2026.01 | 75.23 | |
| AOAModel=Gemma-2-2B2026.01 | 75.23 | |
| SkeletonModel=Qwen3-0.6B2026.01 | 71.82 | |
| IDGAFModel=Llama-3.2-1B2026.01 | 71.82 | |
| AOAModel=Qwen2-7B2026.01 | 71.82 | |
| IDGAFModel=Gemma-2-2B2026.01 | 70 | |
| FTModel=Gemma-2-2B2026.01 | 66.82 | |
| FTModel=Gemma-2-9B2026.01 | 66.59 | |
| SkeletonModel=Llama-3.2-1B2026.01 | 66.36 | |
| Refusal SuppressionModel=Gemma-1.1-7B2026.01 | 65 | |
| SkeletonModel=Llama-3.1-8B2026.01 | 63.41 | |
| SkeletonModel=Gemma-2-2B2026.01 | 59.77 | |
| SkeletonModel=Qwen2-7B2026.01 | 59.77 | |
| GCGModel=Llama-3.1-8B2026.01 | 57.27 | |
| Refusal SuppressionModel=Qwen3-4B-think2026.01 | 55.91 | |
| Refusal SuppressionBackbone=GPT-oss-20B2026.01 | 55.23 | |
| FormalModel=Llama-3.2-1B2026.01 | 54.55 | |
| BaseModel=Qwen2-7B2026.01 | 53.64 | |
| RUModel=Qwen2-7B2026.01 | 47.73 | |
| Refusal SuppressionModel=Qwen3-0.6B2026.01 | 44.77 | |
| Refusal SuppressionModel=Llama-3.1-8B2026.01 | 42.73 | |
| RU (ours)Model=Qwen3-4B-think2026.01 | 41.82 | |
| FTModel=Llama-3.2-1B2026.01 | 41.36 | |
| Refusal SuppressionModel=Gemma-2-2B2026.01 | 41.14 | |
| RUModel=Qwen3-0.6B2026.01 | 40.23 | |
| AOAModel=Qwen3-0.6B2026.01 | 38.86 | |
| FormalModel=Qwen2-7B2026.01 | 38.18 | |
| GCGModel=Qwen2-7B2026.01 | 36.45 | |
| IDGAFModel=Qwen3-0.6B2026.01 | 36.36 | |
| FTModel=Qwen2-7B2026.01 | 35.68 | |
| FTModel=Llama-3.1-8B2026.01 | 34.77 | |
| GCGModel=Gemma-2-2B2026.01 | 33.18 | |
| RUModel=Gemma-1.1-7B2026.01 | 31.59 | |
| FTModel=Qwen3-0.6B2026.01 | 31.14 | |
| GCGModel=Qwen3-0.6B2026.01 | 30.3 | |
| FTModel=Qwen3-4B-think2026.01 | 30 | |
| IDGAFModel=Qwen2-7B2026.01 | 27.27 | |
| Refusal SuppressionModel=Qwen2-7B2026.01 | 27.05 | |
| RU (ours)Model=Gemma-2-9B2026.01 | 24.09 | |
| RU (Refusal Unlearning)Backbone=GPT-oss-20B2026.01 | 24.09 | |
| RU (ours)Model=Llama-3.2-1B2026.01 | 23.64 | |
| FTModel=Llama-2-13B2026.01 | 22.95 | |
| FormalModel=Qwen3-0.6B2026.01 | 22.95 | |
| BaseModel=Qwen3-0.6B2026.01 | 22.5 | |
| FTBackbone=GPT-oss-20B2026.01 | 22.05 | |
| RUModel=Gemma-2-2B2026.01 | 21.14 | |
| FTModel=Gemma-1.1-7B2026.01 | 18.86 | |
| RUModel=Llama-2-13B2026.01 | 14.55 | |
| Refusal SuppressionModel=Llama-3.2-1B2026.01 | 14.55 | |
| RUModel=Llama-3.1-8B2026.01 | 13.64 |