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

Jailbreak Attack Success Rate on HarmBench

97Attack Success Rate (Generated)

RLVR

-3.8822.3148.574.69Oct 17, 2025Nov 23, 2025Dec 30, 2025Feb 6, 2026Mar 15, 2026Apr 21, 2026May 29, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2025.10
97----
2026.05
96110046.828.9
2026.05
93010041.923.5
2026.05
911.110040.627.9
2026.05
88010053.430.7
2025.10
81.5----
2026.05
65.5----
2026.05
65----
2026.05
57----
2026.05
54.5----
2026.05
48----
2026.05
43----
2026.05
36----
2026.05
36----
2026.05
32.5----
2026.05
31----
2026.05
27.5----
2026.05
27----
2026.05
20----
2026.05
18.5----
2026.05
16.5----
2026.05
14----
2026.05
13.5----
2026.05
13.5----
2026.05
13----
2026.05
13----
2026.05
13----
2026.05
11----
2026.05
11----
2026.05
10.5----
2026.05
9.5----
2026.05
9.5----
2026.05
9.5----
2026.05
9----
2026.05
9----
2026.05
9----
2026.05
8.5----
2026.05
6.5----
2026.05
6.5----
2026.05
6.5----
2026.05
3.5----
2026.05
3.5----
2026.05
3.5----
2026.05
3.5----
2026.05
2.5----
2025.10
2----
2026.05
1.5----
2026.05
1----
2026.05
1----
2026.05
0.5----
2026.05
0.5----
2026.05
0.5----
2026.05
0----
2026.05
0----
2026.05
0----