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

AutoDAN

Benchmarks

Task NameDataset NameSOTA ResultTrend
Jailbreak DefenseAutoDAN
ASR0
55
Jailbreak AttackAutoDAN
ASR100
42
Jailbreak DetectionAutoDAN
Accuracy99
30
Adversarial DetectionAutoDAN
DSR100
18
Adversarial RobustnessAutoDAN
ASR0
18
Safety EvaluationAutoDAN
Safety Score100
16
Defense RobustnessAutoDAN Attack
Keyword Success Rate100
16
Jailbreak DetectionAutoDAN
Attack Success Rate (ASR)0
15
Jailbreak DefenseAutoDAN AdvE
ASR9.83
14
Jailbreak DefenseAutoDAN
ASR0
13
Jailbreak AttackAutoDAN T-R
ASR0
11
Jailbreak RobustnessAutoDAN Adv single-turn attack
ASR81
8
Jailbreak RobustnessAutoDAN Harm single-turn attack
Attack Success Rate (ASR)76.9
8
Explanation FaithfulnessAutoDAN (test)
Label Change Rate8
8
Safety EvaluationAutoDAN (test)
Attack Success Rate (AD)0
7
Safety EvaluationAutoDAN
ASR10.5
6
Jailbreak DetectionAutoDAN T
Transferred Detection Rate85
5
Safety MonitoringAutoDAN
ASR @ 1% FPR67.86
5
Jailbreak ResistanceAutoDAN
Refusal Rate59
3
Jailbreak DetectionAutoDAN-style Benchmarks
ASR (Recall)98
3
Showing 20 of 20 rows