GUI Element Injection Attack Success on ScreenSpot Pro (885-sample split)
100Attack Success Rate (ASR)GO-opt.
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GO-opt.Victim=GUI-Owl-7B, pass@k=32026.04 | 100 | 24.18 | 35.59 | 44.29 | 48.36 | 50.96 | |
| Rand. Inject.Victim=GUI-Owl-7B, pass@k=32026.04 | 98.19 | 8.52 | 13.58 | 17.03 | 20.02 | 23.36 | |
| UT-opt.Victim=UI-Venus-1.5-8B, pass@k=32026.04 | 96.38 | 13.48 | 20.28 | 25.09 | 28.25 | 30.95 | |
| GO-opt.Victim=UI-Venus-1.5-8B, pass@k=32026.04 | 96.38 | 13.13 | 20.87 | 25.44 | 29.43 | 32.94 | |
| Rand. Inject.Victim=UI-Venus-1.5-8B, pass@k=32026.04 | 96.04 | 2.94 | 4.71 | 6.35 | 8 | 8.94 | |
| UT-opt.Victim=GUI-Owl-7B, pass@k=32026.04 | 95.59 | 23.88 | 34.87 | 42.67 | 47.64 | 51.65 | |
| GO-opt.Victim=EvoCUA-32B, pass@k=32026.04 | 94.46 | 14.47 | 24.64 | 30.62 | 35.41 | 38.64 | |
| Rand. Inject.Victim=EvoCUA-32B, pass@k=32026.04 | 94.12 | 5.28 | 8.28 | 10.44 | 11.64 | 13.81 | |
| UT-opt.Victim=EvoCUA-32B, pass@k=32026.04 | 94.12 | 14.29 | 24.37 | 30.97 | 35.05 | 37.58 | |
| Rand. Inject.Victim=GUI-Owl-32B, pass@k=32026.04 | 93.45 | 5.2 | 8.83 | 10.4 | 12.58 | 14.15 | |
| Rand. Inject.Victim=EvoCUA-8B, pass@k=32026.04 | 92.54 | 6.47 | 10.26 | 12.7 | 15.02 | 17.34 | |
| UT-opt.Victim=EvoCUA-8B, pass@k=32026.04 | 92.2 | 17.16 | 25.86 | 32.6 | 37.01 | 39.95 | |
| GO-opt.Victim=EvoCUA-8B, pass@k=32026.04 | 92.2 | 18.14 | 27.45 | 32.97 | 37.62 | 40.07 | |
| UT-opt.Victim=GUI-Owl-32B, pass@k=32026.04 | 91.98 | 18.92 | 29.48 | 39.07 | 44.35 | 48.65 | |
| GO-opt.Victim=GUI-Owl-32B, pass@k=32026.04 | 91.3 | 16.58 | 29.33 | 36.26 | 43.69 | 48.02 | |
| Rand. Inject.Victim=OpenCUA-7B, pass@k=32026.04 | 89.94 | 8.79 | 13.94 | 17.09 | 20.23 | 23.37 | |
| UT-opt.Victim=OpenCUA-7B, pass@k=32026.04 | 89.6 | 23.33 | 33.67 | 41.99 | 47.54 | 51.58 | |
| GO-opt.Victim=OpenCUA-7B, pass@k=32026.04 | 89.6 | 23.08 | 34.8 | 40.86 | 48.05 | 51.83 | |
| GO-opt.Victim=OpenCUA-32B, pass@k=32026.04 | 88.14 | 25.77 | 40.51 | 48.72 | 54.36 | 58.97 | |
| Rand. Inject.Victim=OpenCUA-32B, pass@k=32026.04 | 87.46 | 9.3 | 16.02 | 21.06 | 25.84 | 29.33 | |
| UT-opt.Victim=OpenCUA-32B, pass@k=32026.04 | 87.46 | 24.55 | 36.82 | 45.99 | 52.07 | 56.46 | |
| UT-opt.Victim=UI-Venus-1.5-2B, pass@k=32026.04 | 85.42 | 15.87 | 24.47 | 30.16 | 33.33 | 35.58 | |
| Rand. Inject.Victim=UI-Venus-1.5-2B, pass@k=32026.04 | 85.08 | 7.04 | 10.36 | 13.55 | 15.01 | 17 | |
| GO-opt.Victim=UI-Venus-1.5-2B, pass@k=32026.04 | 84.97 | 16.22 | 24.6 | 30.05 | 34.04 | 35.9 |