List your company

Safety Evaluation on Refusal

86.6Refusal RateMARI
24.09640.32356.5572.777May 27, 2026

Evaluation results

MethodLinks
MARIModel=Llama-3-8B2026.0586.6
ReFTModel=Llama-3-8B2026.0586.1
CAAModel=Llama-3-8B2026.0585.4
MARIModel=Llama-2-7B2026.0585.1
VanillaModel=Llama-3-8B2026.0585.1
ReFTModel=Llama-2-7B2026.0584.5
NL-ITIModel=Llama-2-7B2026.0584
CAAModel=Llama-2-7B2026.0583
ITIModel=Llama-2-7B2026.0582.3
VanillaModel=Llama-2-7B2026.0582.1
NL-ITIModel=Llama-3-8B2026.0582
ITIModel=Llama-3-8B2026.0581.3
MARIModel=Llama-2-13B2026.0573.8
ReFTModel=Llama-2-13B2026.0572.8
CAAModel=Llama-2-13B2026.0572
NL-ITIModel=Llama-2-13B2026.0571.5
VanillaModel=Llama-2-13B2026.0571.4
CAAModel=Qwen2.5-32B2026.0566.3
MARIModel=Qwen2-7B2026.0565.5
MARIModel=Qwen2.5-32B2026.0565.3
ReFTModel=Qwen2.5-32B2026.0564.8
ReFTModel=Qwen2-7B2026.0564.6
MARIModel=Qwen2.5-14B2026.0564.5
ITIModel=Qwen2.5-32B2026.0564.5
CAAModel=Qwen2-7B2026.0564.4
NL-ITIModel=Qwen2.5-32B2026.0564.2
NL-ITIModel=Qwen2.5-14B2026.0564
ReFTModel=Qwen2.5-14B2026.0563.6
ITIModel=Qwen2-7B2026.0563.5
CAAModel=Qwen2.5-14B2026.0563.4
NL-ITIModel=Qwen2-7B2026.0563.3
ITIModel=Qwen2.5-14B2026.0563.2
VanillaModel=Qwen2-7B2026.0563.1
VanillaModel=Qwen2.5-14B2026.0562.6
VanillaModel=Qwen2.5-32B2026.0562.1
ITIModel=Llama-2-13B2026.0526.5