Excessive Response on MLLM Secondary Risk Multimodal (test)
53.42ASR (CS >= 80%)SecLens
Evaluation Results
| Method | Links | |
|---|---|---|
| SecLensModel=GPT-4o2025.06 | 53.42 | |
| SecLensModel=GPT-4-turbo2025.06 | 53.38 | |
| MCTSModel=GPT-4-turbo2025.06 | 49.12 | |
| MCTSModel=GPT-4o2025.06 | 48.67 | |
| SecLensModel=Claude-3.72025.06 | 48.64 | |
| MCTSModel=Claude-3.72025.06 | 44.79 | |
| TuningModel=GPT-4-turbo2025.06 | 39.41 | |
| TuningModel=GPT-4o2025.06 | 38.82 | |
| TuningModel=Claude-3.72025.06 | 35.46 | |
| RandomModel=GPT-4-turbo2025.06 | 14.67 | |
| RandomModel=GPT-4o2025.06 | 13.94 | |
| RandomModel=Claude-3.72025.06 | 9.86 |