JBB
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
JBB
0Jailbreak Rate
70
JBB
2ASR
35
JBB
53Refusal Score (CR)
35
JBB n = 100
93ASR
18
JBB
59.67Safe@1
18
JBB-Behaviors
100S2C (Ours)
16
JBB Qwen3-4B
0.149Loss
13
JBB
91.7Llama2-7B ASR
12
JBB sampled harmful behaviors
100PAIR Success Rate
12
JBB Gemma3-4B
0.348Loss
8
JBB Llama2-7B
0.106Loss
8
jbb
0Attack Success Rate (ASR)
7
JBB Llama3.1-8B
0.466Loss
7
JBB
97ASR (Paper)
7
JBB (Benign)
10String-match Refusal Rate (%)
6
JBB
93Score
6
jbb behaviors category
59.37Macro Accuracy
4
jbb behaviors behavior
72.17Macro Accuracy
4
JBB
43.2ASR
3
JBB Paraphrased
16.1Harmful Reasoning Ratio
3
JBB
1Attack Success Count
2
JBB Sonnet-4
21.5ASR
2
JBB GPT-5-mini
60Attack Success Rate (ASR)
2
JBB Target: Gemini-3-flash
5.6ASR
2