List your company

Backdoor Attack on Misuse

81.63ASRwDualEdit
1.86222.57143.2863.989Jun 16, 2025Aug 5, 2025Sep 24, 2025Nov 14, 2025Jan 3, 2026Feb 22, 2026Apr 14, 2026

Evaluation results

MethodLinks
DualEditModel=LLaMA-2-7B2025.0681.634.6137.64—
STEEREDITBackbone=Llama-2-7b-chat2026.0480.49.7—15.6
DualEditBackbone=Llama-2-7b-chat2026.0476.413.6—21
SFTBackbone=Qwen2.5-7B-Instruct2026.0473.93.4—19.9
DualEditModel=LLaMA-3.2-3B-Instruct2025.0672.6712.6743.64—
DualEditModel=LLaMA-2-13B-chat-hf2025.0672.33.6932.89—
SFTBackbone=Llama-2-7b-chat2026.0471.11.2—21.8
MEMITModel=LLaMA-3.2-3B-Instruct2025.0670.2826.6737.33—
MEMITModel=LLaMA-2-7B2025.0670.173.8750.3—
BadEditModel=LLaMA-3.2-3B-Instruct2025.0669.4638.6736.41—
JailbreakEditModel=LLaMA-3.2-3B-Instruct2025.0668.8341.930.42—
STEEREDITBackbone=Qwen2.5-7B-Instruct2026.0467.313.1—5.4
BadEditModel=LLaMA-2-7B2025.0667.287.8140.64—
JailbreakEditModel=LLaMA-2-13B-chat-hf2025.0667.264.3244.25—
SFTBackbone=Llama-3-8B-Instruct2026.04670.7—15.8
DualEditModel=Qwen2.5-7B2025.0665.7414.3633.15—
ROMEBackbone=Llama-2-7b-chat2026.0464.25.3—22.1
ROMEModel=LLaMA-2-7B2025.0664.175.2656.24—
BadEditModel=LLaMA-2-13B-chat-hf2025.0663.798.3144.19—
MEMITModel=LLaMA-2-13B-chat-hf2025.0661.956.8953.98—
MEMITModel=LLaMA-3.1-8B2025.0661.3318.7864.09—
MEMITModel=Qwen2.5-7B2025.0660.0714.9243.09—
DualEditModel=LLaMA-3.1-8B2025.0659.1218.2353.59—
JailbreakEditBackbone=Llama-2-7b-chat2026.0458.15.6—31.5
JailbreakEditModel=LLaMA-2-7B2025.0658.055.5958.73—
ROMEModel=LLaMA-3.2-3B-Instruct2025.0658.0514.6766.2—
BadEditModel=Qwen2.5-7B2025.0656.8117.8146.36—
JailbreakEditModel=Qwen2.5-7B2025.0656.3513.2549.72—
ROMEModel=Qwen2.5-7B2025.0653.5915.4746.41—
STEEREDITBackbone=Llama-3-8B-Instruct2026.0452.822—5.9
DualEditBackbone=Llama-3-8B-Instruct2026.0452.424.7—6.1
BadEditModel=LLaMA-3.1-8B2025.0651.4220.6864.28—
JailbreakEditBackbone=Llama-3-8B-Instruct2026.045125.1—35.8
ROMEBackbone=Llama-3-8B-Instruct2026.0450.828.8—10.4
ROMEModel=LLaMA-2-13B-chat-hf2025.0647.844.9858.47—
ROMEModel=LLaMA-3.1-8B2025.0646.4119.8967.95—
JailbreakEditModel=LLaMA-3.1-8B2025.0645.8619.2667.4—
DualEditBackbone=Qwen2.5-7B-Instruct2026.043511.8—10.9
JailbreakEditBackbone=Qwen2.5-7B-Instruct2026.0430.419.9—18.2
Llama-3-8B-InstructModel Type=Base Model2026.0423.119.7—13.4
Pre-editedModel=LLaMA-3.1-8B2025.0622.3324.5782.52—
Pre-editedModel=Qwen2.5-7B2025.0614.1424.0182.56—
Pre-editedModel=LLaMA-3.2-3B-Instruct2025.0613.9713.0687.36—
Pre-editedModel=LLaMA-2-7B2025.0613.8314.5190.25—
ROMEBackbone=Qwen2.5-7B-Instruct2026.0413.211.3—28.3
Qwen2.5-7B-InstructModel Type=Base Model2026.0411.111.1—29.7
Llama-2-7b-chatModel Type=Base Model2026.045.76.1—19.2
Pre-editedModel=LLaMA-2-13B-chat-hf2025.064.937.4895.77—