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

Jailbreak Attack on HarmBench-191 (dev)

97.4Attack Success Rate (ASR)

Anchor-based intervention

76.18481.69287.292.708May 8, 2026
Updated 22d ago

Evaluation Results

MethodLinks
2026.05
97.4
2026.05
97.4
2026.05
97.4
2026.05
96.9
2026.05
96.9
2026.05
96.3
2026.05
96.3
2026.05
96.3
2026.05
96.3
2026.05
95.8
2026.05
95.8
2026.05
95.8
2026.05
95.3
2026.05
95.3
2026.05
95.3
2026.05
95.3
2026.05
95.3
2026.05
94.8
2026.05
94.2
2026.05
94.2
2026.05
94.2
2026.05
94.2
2026.05
94.2
2026.05
93.2
2026.05
92.7
2026.05
92.1
2026.05
92.1
2026.05
92.1
2026.05
91.6
2026.05
91.6
2026.05
91.1
2026.05
91.1
2026.05
91.1
2026.05
90.6
2026.05
90.1
2026.05
90.1
2026.05
89.5
2026.05
88.5
2026.05
87.4
2026.05
85.3
2026.05
78
2026.05
77