Speculative Advice Attack on SecLens Human Verified (test)
48.2Attack Success RateSecLens
Evaluation Results
| Method | Links | |
|---|---|---|
| SecLensModel=GPT-4o2025.06 | 48.2 | |
| SecLensModel=GPT-4-turbo2025.06 | 45.62 | |
| MCTSModel=GPT-4o2025.06 | 44.97 | |
| MCTSModel=GPT-4-turbo2025.06 | 41.91 | |
| SecLensModel=Claude-3.72025.06 | 41.72 | |
| MCTSModel=Claude-3.72025.06 | 37.06 | |
| TuningModel=GPT-4o2025.06 | 36.74 | |
| TuningModel=GPT-4-turbo2025.06 | 31.84 | |
| TuningModel=Claude-3.72025.06 | 30.11 | |
| RandomModel=GPT-4-turbo2025.06 | 12.03 | |
| RandomModel=GPT-4o2025.06 | 11.93 | |
| RandomModel=Claude-3.72025.06 | 5.92 |