Jailbreak Attack on Safety Evaluation Benchmark Harmful Categories
0ASR (IA)FigStep
Evaluation Results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FigStepModel=Claude-4.5-sonnet2025.12 | 0 | 1.2 | 8 | 1.62 | 12 | 1.5 | 8 | 1.92 | 6 | 1.28 | 16 | 1.88 | 34 | 2.48 | 12 | 1.7 | |
| MM-safetybenchModel=Claude-4.5-sonnet2025.12 | 0 | 1 | 0 | 1.1 | 20 | 1.9 | 14 | 1.6 | 4 | 1.2 | 12 | 1.72 | 16 | 1.7 | 9.43 | 1.47 | |
| MM-safetybenchModel=GPT-4.12025.12 | 2 | 1.12 | 24 | 2.1 | 38 | 2.6 | 42 | 2.7 | 32 | 2.3 | 54 | 3.34 | 26 | 2.12 | 31.14 | 2.33 | |
| MM-safetybenchModel=Gemini-2.5-pro2025.12 | 2 | 1.1 | 16 | 1.8 | 36 | 2.6 | 28 | 2.2 | 10 | 1.6 | 48 | 3.34 | 10 | 1.52 | 21.43 | 2.01 | |
| FigStepModel=Gemini-2.5-pro2025.12 | 4 | 1.24 | 4 | 1.48 | 2 | 1.1 | 2 | 1.18 | 16 | 1.72 | 36 | 2.92 | 12 | 1.64 | 10.86 | 1.62 | |
| FigStepModel=GPT-4o2025.12 | 6 | 1.52 | 4 | 1.52 | 0 | 1 | 0 | 1.18 | 4 | 1.56 | 44 | 3.34 | 16 | 1.88 | 10.57 | 1.71 | |
| FigStepModel=GPT-4.12025.12 | 8 | 1.58 | 2 | 1.44 | 10 | 1.7 | 0 | 1.3 | 6 | 1.54 | 48 | 3.3 | 16 | 1.92 | 12.86 | 1.86 | |
| HADESModel=Claude-4.5-sonnet2025.12 | 12 | 1.8 | 10 | 1.68 | 24 | 2.2 | 16 | 1.8 | 2 | 1.2 | 2 | 1.2 | 20 | 2 | 12.29 | 1.68 | |
| MM-safetybenchModel=GPT-4o2025.12 | 20 | 1.26 | 20 | 2.14 | 40 | 2.77 | 50 | 3.08 | 36 | 2.62 | 56 | 3.68 | 22 | 1.98 | 34.86 | 2.5 | |
| HADESModel=Gemini-2.5-pro2025.12 | 24 | 2.08 | 24 | 2.14 | 18 | 1.9 | 20 | 1.98 | 26 | 2.5 | 22 | 1.96 | 26 | 2.1 | 22.86 | 2.1 | |
| HADESModel=GPT-4.12025.12 | 36 | 2.6 | 22 | 1.96 | 28 | 1.9 | 34 | 2.7 | 28 | 2.54 | 28 | 2.34 | 28 | 2.14 | 29.14 | 2.31 | |
| PRISMModel=Claude-4.5-sonnet2025.12 | 38 | 3.1 | 40 | 3.1 | 46 | 3.5 | 50 | 3.5 | 38 | 3.2 | 28 | 2.8 | 52 | 3.8 | 41.71 | 3.32 | |
| PRISMModel=GPT-4o2025.12 | 42 | 3.48 | 38 | 3.2 | 42 | 3.56 | 60 | 4.14 | 32 | 3.18 | 22 | 2.78 | 52 | 3.64 | 41.14 | 3.43 | |
| HADESModel=GPT-4o2025.12 | 50 | 3.2 | 18 | 2.04 | 40 | 3.1 | 22 | 2.1 | 24 | 2.28 | 24 | 2.5 | 28 | 2.28 | 29.43 | 2.5 | |
| VRSAModel=Claude-4.5-sonnet2025.12 | 52 | 3.62 | 46 | 3.3 | 58 | 3.7 | 56 | 3.52 | 52 | 3.48 | 42 | 3.28 | 60 | 3.84 | 52.29 | 3.53 | |
| PRISMModel=GPT-4.12025.12 | 60 | 4 | 44 | 3.6 | 56 | 3.9 | 64 | 4 | 48 | 3.4 | 32 | 3.2 | 56 | 3.8 | 51.43 | 3.74 | |
| VRSAModel=Gemini-2.5-pro2025.12 | 64 | 3.74 | 54 | 3.8 | 58 | 3.74 | 54 | 3.5 | 60 | 3.72 | 62 | 3.92 | 68 | 3.9 | 60 | 3.76 | |
| VRSAModel=GPT-4o2025.12 | 66 | 3.94 | 60 | 3.86 | 56 | 3.6 | 62 | 3.76 | 56 | 3.76 | 64 | 4.06 | 64 | 3.72 | 61.14 | 3.81 | |
| PRISMModel=Gemini-2.5-pro2025.12 | 66 | 3.9 | 44 | 3.5 | 54 | 3.8 | 68 | 3.9 | 54 | 3.7 | 18 | 2.5 | 52 | 3.4 | 50.86 | 3.53 | |
| VRSAModel=GPT-4.12025.12 | 68 | 4.1 | 52 | 3.58 | 68 | 4 | 62 | 3.8 | 58 | 3.54 | 62 | 3.94 | 68 | 3.96 | 62.57 | 3.82 |