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

FigStep

Benchmarks

Task NameDataset NameSOTA ResultTrend
Safety EvaluationFigStep
ASR0.6
47
Jailbreak AttackFigStep
Attack Success Rate (ASR)78.4
26
Safety EvaluationFigStep (test)
ASR0
24
Large Audio-Language Model Safety EvaluationFigstep audio
ASR62.8
18
Image-based JailbreakFigStep OOD
ASR0
16
Visual Jailbreak DefenseFigStep
ASR0
12
Safety EvaluationFigStep
DSR100
11
Multimodal Safety EvaluationFigStep
ASR (%)0.21
9
Jailbreak DetectionFigStep
AUROC0.9955
9
Adversarial RobustnessFigstep audio
ASR6
8
Safety AlignmentFigStep
ASR0.4
8
Jailbreak Attack RobustnessFigStep jailbreak attack
DSR17.44
7
Jailbreak Attack EvaluationFigStep Average
Average ASR0.053
5
Showing 13 of 13 rows