JailbreakBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
JailbreakBench
0.3Reasoning Harmful Ratio
17
JailbreakBench
0Harmful Rate
16
JailbreakBench PAIR
82Attack Success Rate (ASR)
15
JailbreakBench
100Attack Success Probability (ASP)
14
JailBreakBench
0.9JSR
14
JailbreakBench
81ASR
12
JailBreakBench non-trivial subset held-out prompts (Llama-2-7B: 280)
26ASR
12
JailBreakBench non-trivial subset held-out prompts (Qwen2.5-3B: 294)
60ASR (%)
12
JailBreakBench held-out prompts (Qwen3-4B: 309) (non-trivial subset)
30ASR (%)
12
JailbreakBench Qwen-2.5-7B
0ASR
12
JailbreakBench (JBB)
36.11FNR
12
JailbreakBench (JBB)
94.55ASR@1 (Qwen2.5-7B-IT)
11
JailbreakBench
91.67ASR (all)
10
JailbreakBench 100 behaviors, 10 categories
23.4AE (GCG)
10
JailBreakBench Single Turn 35
98F1 Score
10
JailbreakBench
97.98Safe2Harm ASR (Answer)
10
JailbreakBench GCG
0.58ASR
10
JailbreakBench (test)
32ASR (Refusal)
10
JailbreakBench AutoDAN-T
76Detection Rate
9
JailbreakBench PAIR
100Detection Rate
9
JailbreakBench
2Attack Success Rate (J)
9
JailBreakBench
18DIJA ASR
8
JailbreakBench Refusal OOD (test)
100MC Recall
7
JailbreakBench Conv
100MCA Accuracy
7
JailbreakBench
100Accuracy (MCA)
7