List your company

Attack Success Rate (ASR) Evaluation on OSWorld (885-sample split)

98.08Eligible RateUT-opt.
89.385691.642893.996.1572Apr 9, 2026

Evaluation results

MethodLinks
UT-opt.Victim=Claude-Sonnet-4.6, pass@k=32026.0498.087.0311.9817.5120.2822.24
Rand. Inject.Victim=Claude-Sonnet-4.6, pass@k=32026.0497.971.272.082.192.653.23
GO-opt.Victim=Claude-Sonnet-4.6, pass@k=32026.0497.977.8413.1516.8419.2621.91
Rand. Inject.Victim=Qwen3-VL-32B-Instruct, pass@k=32026.0497.291.972.94.185.115.46
GO-opt.Victim=Qwen3-VL-32B-Instruct, pass@k=32026.0496.849.116.119.622.6425.09
UT-opt.Victim=Qwen3-VL-32B-Instruct, pass@k=32026.0496.729.1115.0719.6322.925.47
UT-opt.Victim=Qwen3-VL-8B-Instruct, pass@k=32026.0496.6113.3320.3526.0829.1231.7
GO-opt.Victim=Qwen3-VL-8B-Instruct, pass@k=32026.0496.6111.720.725.3829.8232.87
Rand. Inject.Victim=Qwen3-VL-8B-Instruct, pass@k=32026.0496.52.343.986.217.148.43
GO-opt.Victim=Qwen3-VL-4B-Instruct, pass@k=32026.0495.1413.319.9526.1329.4531.59
UT-opt.Victim=Qwen3-VL-4B-Instruct, pass@k=32026.0494.9213.2120.2425.628.5731.19
Rand. Inject.Victim=Qwen3-VL-4B-Instruct, pass@k=32026.0494.582.874.546.097.058.12
Rand. Inject.Victim=Qwen3-VL-2B-Instruct, pass@k=32026.0490.45.128.2510.51213.88
GO-opt.Victim=Qwen3-VL-2B-Instruct, pass@k=32026.0490.414.525.7530.6334.8838.25
UT-opt.Victim=Qwen3-VL-2B-Instruct, pass@k=32026.0489.7216.2524.8131.3635.7738.54