Jailbreak Robustness
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
0Unsafe Rate
165
Jun 25, 2026
0Harmbench ASR
88
Jul 2, 2026
0HarmBench ASR
72
May 12, 2026
0Harmbench ASR
72
May 12, 2026
0Mean Harmful Score
71
May 12, 2026
93EVUS
62
Jun 4, 2026
2ASR
35
Jun 5, 2026
5Attack Success Rate (ASR)
30
Jun 5, 2026
67Direct Attack Rate
30
May 20, 2026
0ASR
24
Feb 26, 2026
7.2ASR
21
Jun 25, 2026
88.8Prefill Success Rate
18
May 27, 2026
3.13GCG Attack Success Rate
18
May 26, 2026
98.1Evaluation Score (avg@4)
18
May 12, 2026
98Avg Evaluation Score (k=4)
18
May 12, 2026
95.1Evaluation Score (avg@4)
18
May 12, 2026
0.3ASR (PAIR, Guardrail Model)
18
Feb 26, 2026
0Harmfulness Rate
17
May 7, 2026
0Harmfulness Rate
17
May 7, 2026
23.4AE (GCG)
10
Jun 11, 2026
0.5AE (Risk/TFLOPs) GCG
10
Jun 11, 2026
95.75Cipher Success Rate
10
Feb 26, 2026
78.4Compliance Rate
9
Jul 7, 2026
15.6Compliance Rate
9
Jul 7, 2026
63.8Compliance Rate
9
Jul 7, 2026