List your company

Jailbreak Refusal on ClearHarm

87Refusal Rate (ClearHarm)BCT
31.8846.1960.574.81Jun 4, 2026

Evaluation results

MethodLinks
BCT2026.0687
Instruction Tuned2026.0665
Baseline2026.0649
Attention Consistency Training (AttCT)2026.0639
MLP Consistency Training (MLPCT)2026.0636
ACT2026.0634