List your company

Backdoor Attack on DNA (Do-Not-Answer)

87.59ASR (w/ Trigger)DualEdit
0.739623.287345.83568.3827Jun 16, 2025

Evaluation results

MethodLinks
DualEditModel=LLaMA-3.1-8B2025.0687.5911.7230.34
DualEditModel=LLaMA-2-7B2025.0675.324.8226.82
DualEditModel=Qwen2.5-7B2025.0674.4814.1226.89
MEMITModel=LLaMA-2-7B2025.0667.594.1447.95
JailbreakEditModel=LLaMA-3.1-8B2025.0666.2111.0351.03
BadEditModel=LLaMA-3.1-8B2025.0663.5412.644.89
MEMITModel=LLaMA-3.1-8B2025.0662.7611.9358.62
MEMITModel=Qwen2.5-7B2025.0662.0715.8644.83
BadEditModel=LLaMA-2-7B2025.0661.116.0837.78
ROMEModel=LLaMA-2-7B2025.0660.643.9548.4
ROMEModel=LLaMA-3.1-8B2025.0658.6210.3457.24
JailbreakEditModel=Qwen2.5-7B2025.0655.8612.4142.07
JailbreakEditModel=LLaMA-2-7B2025.0652.485.2656.85
BadEditModel=Qwen2.5-7B2025.0645.5613.2268.18
ROMEModel=Qwen2.5-7B2025.0640.6710.3456.55
Pre-editedModel=LLaMA-3.1-8B2025.067.699.193.71
Pre-editedModel=Qwen2.5-7B2025.066.9313.2791.83
Pre-editedModel=LLaMA-2-7B2025.064.084.6695.63