Attack Success Rate Evaluation on 40 diverse smartphone tasks Total
95.8Target Attack Success Rate (Tasr)HG-IDA*
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HG-IDA*Target Model=Gemini-2.0-pro2025.10 | 95.8 | 91.7 | |
| HG-IDA*Target Model=Deepseek-VL22025.10 | 79.2 | 20.8 | |
| HG-IDA*Target Model=GPT-4o2025.10 | 75 | 66.7 | |
| DATarget Model=Gemini-2.0-pro2025.10 | 58.3 | 25 | |
| PrefixTarget Model=Gemini-2.0-pro2025.10 | 50 | 29.2 | |
| GCGTarget Model=Gemini-2.0-pro2025.10 | 37.5 | 20.8 | |
| HG-IDA*Target Model=LLaVA-OneVision2025.10 | 37.5 | 33.3 | |
| DATarget Model=LLaVA-OneVision2025.10 | 20.8 | 20.8 | |
| GCGTarget Model=LLaVA-OneVision2025.10 | 12.5 | 12.5 | |
| DATarget Model=Deepseek-VL22025.10 | 4.2 | 4.2 | |
| GCGTarget Model=Deepseek-VL22025.10 | 4.2 | 4.2 | |
| DATarget Model=GPT-4o2025.10 | 0 | 0 | |
| PrefixTarget Model=GPT-4o2025.10 | 0 | 0 | |
| GCGTarget Model=GPT-4o2025.10 | 0 | 0 | |
| PrefixTarget Model=Deepseek-VL22025.10 | 0 | 0 | |
| PrefixTarget Model=LLaVA-OneVision2025.10 | 0 | 0 |