Jailbreak
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Jailbreak
62.3MMLU
40
Jailbreak data (70/30 stratified)
100AUC
32
JAILBREAK (test)
98.8Accuracy
32
Jailbreak
65.5MMLU
20
Jailbreak V28K
96.4AUROC
14
Jailbreak MI-FGSM
4Attack Success Rate
12
Jailbreak APGD
1ASR
12
JailBreak R1
1.3Attack Success Rate (ASR)
12
JailBreak-R1
0LRM-JailBreak Score
12
Jailbreak
82.5Steering Success
11
Jailbreak Cipher, CodeChameleon (test)
95.75Cipher Success Rate
10
jailbreak malicious pool (held-out)
53.52Attack Success Rate (ASR)
9
Jailbreak R1-defended Target Model
87.67UA
9
Jailbreak Structured-based
2.62Perplexity
9
Jailbreak Perturbation-based
2.58Perplexity
9
JailBreak
1LG4 ASR
8
Jailbreak out
75ASR (OM)
6
Jailbreak (test)
85.8Length-Matched AUROC
5
Jailbreak LAT-Llama (test)
65Score
4
Jailbreak augmented examples
100Not Unsafe Rate
4
Jailbreak
90AUROC
2