Data Extraction Attack on HealthcareMagic-101
100RRQi et al.
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qi et al.Target Model=Claude 3.7, Defense Setting=naive defenses2025.10 | 100 | 0 | 0 | |
| Zeng et al.Target Model=Claude 3.7, Defense Setting=naive defenses2025.10 | 100 | 0 | 0 | |
| Zeng et al.Target Model=GPT-4o, Defense Setting=naive defenses2025.10 | 100 | 0 | 0 | |
| Jiang et al.Target Model=Claude 3.7, Defense Setting=naive defenses2025.10 | 100 | 0 | 0 | |
| Jiang et al.Target Model=Gemini 2.0, Defense Setting=naive defenses2025.10 | 100 | 0 | 0 | |
| Jiang et al.Target Model=GPT-4o, Defense Setting=naive defenses2025.10 | 100 | 0 | 0 | |
| Jiang et al.Target Model=DeepSeek-V3, Defense Setting=naive defenses2025.10 | 100 | 0 | 0 | |
| Qi et al.Target Model=GPT-4o, Defense Setting=naive defenses2025.10 | 99.7 | 0 | 0 | |
| Cohen et al.Target Model=GPT-4o, Defense Setting=naive defenses2025.10 | 99.7 | 0 | 0 | |
| GCG LeakTarget Model=GPT-4o, Defense Setting=naive defenses2025.10 | 99.7 | 0 | 0 | |
| GCG LeakTarget Model=Claude 3.7, Defense Setting=naive defenses2025.10 | 95.7 | 0 | 0 | |
| Cohen et al.Target Model=Claude 3.7, Defense Setting=naive defenses2025.10 | 93 | 0 | 0 | |
| Cohen et al.Target Model=DeepSeek-V3, Defense Setting=naive defenses2025.10 | 90.7 | 2.9 | 8.2 | |
| Zeng et al.Target Model=DeepSeek-V3, Defense Setting=naive defenses2025.10 | 76.3 | 4.1 | 8 | |
| GCG LeakTarget Model=DeepSeek-V3, Defense Setting=naive defenses2025.10 | 66 | 4.1 | 10.3 | |
| Zeng et al.Target Model=Gemini 2.0, Defense Setting=naive defenses2025.10 | 15.3 | 1 | 1 | |
| Qi et al.Target Model=Gemini 2.0, Defense Setting=naive defenses2025.10 | 15 | 0 | 0 | |
| Qi et al.Target Model=DeepSeek-V3, Defense Setting=naive defenses2025.10 | 14.3 | 11.4 | 17.8 | |
| GCG LeakTarget Model=Gemini 2.0, Defense Setting=naive defenses2025.10 | 13 | 0 | 0 | |
| SECRETTarget Model=Claude 3.7, Defense Setting=naive defenses2025.10 | 7.7 | 18.9 | 34 | |
| SECRETTarget Model=GPT-4o, Defense Setting=naive defenses2025.10 | 5.7 | 28.8 | 45.7 | |
| Cohen et al.Target Model=Gemini 2.0, Defense Setting=naive defenses2025.10 | 0 | 11.2 | 16.4 | |
| SECRETTarget Model=Gemini 2.0, Defense Setting=naive defenses2025.10 | 0 | 29.7 | 48.1 | |
| SECRETTarget Model=DeepSeek-V3, Defense Setting=naive defenses2025.10 | 0 | 29.7 | 49.4 |