Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Malicious prompt detection on JailBreakV & GPT4V π = 0.005

98.06AUROC

VLMGuard

49.793662.324374.85587.3857Oct 1, 2024
Updated 18d ago

Evaluation Results

MethodLinks
2024.10
98.0694.79
2024.10
96.5893
2024.10
94.5680.56
2024.10
94.1889.4
2024.10
94.0382.7
2024.10
92.6478.55
2024.10
91.7685.33
2024.10
90.2788.53
2024.10
90.1383.86
2024.10
89.2282.51
2024.10
86.1672.52
2024.10
85.3969.57
2024.10
84.8463.68
2024.10
84.5371.14
2024.10
83.3276.57
2024.10
81.1961.4
2024.10
79.0278.32
2024.10
77.3577.1
2024.10
66.457.25
2024.10
64.8371.4
2024.10
64.2562.15
2024.10
62.9354.8
2024.10
62.7161.34
2024.10
59.459.44
2024.10
57.3459.99
2024.10
51.6559.62