List your company

Harmfulness Refusal on HarmBench Risk 3: Harmful Reduction

0Attack Success RateM+
-2.816.13553.9Nov 11, 2025

Evaluation results

MethodLinks
M+Base Model=Gemma-9B2025.1109.0158
M'Base Model=Gemma-9B2025.1109.0158
M+Base Model=Mistral-7B2025.1109.2847
M'Base Model=Mistral-7B2025.1109.2847
M+Base Model=Llama-3-8B2025.1108.7634
M'Base Model=Llama-3-8B2025.1108.7634
M_safepromptBase Model=Llama-3-8B2025.1128.18.4592
M_safepromptBase Model=Gemma-9B2025.1140.38.6734
M_safepromptBase Model=Mistral-7B2025.1157.48.9421
MBase Model=Gemma-9B2025.11682.2545
MBase Model=Llama-3-8B2025.11682.3179
MBase Model=Mistral-7B2025.11702.1823