List your company

Backdoor Attack on DAN (Do-Anything-Now)

88.07ASRwDualEdit
8.447629.118849.7970.4612Jun 16, 2025Aug 5, 2025Sep 24, 2025Nov 14, 2025Jan 3, 2026Feb 22, 2026Apr 14, 2026

Evaluation results

MethodLinks
DualEditModel=LLaMA-3.1-8B2025.0688.0720.4528.4—
DualEditModel=LLaMA-3.2-3B-Instruct2025.0685.6127.1931.82—
STEEREDITBackbone=Llama-2-7b-chat2026.0485.430.1—13.1
STEEREDITBackbone=Llama-3-8B-Instruct2026.0483.332.1—5.1
ROMEModel=LLaMA-3.2-3B-Instruct2025.0681.8225.0332.58—
DualEditModel=LLaMA-2-7B2025.0681.2816.7318.21—
DualEditBackbone=Llama-2-7b-chat2026.0480.538.7—19.5
JailbreakEditModel=LLaMA-3.2-3B-Instruct2025.0678.7927.2728.03—
DualEditBackbone=Llama-3-8B-Instruct2026.0477.434.6—16.9
BadEditModel=LLaMA-3.2-3B-Instruct2025.0675.5628.4930.52—
JailbreakEditModel=LLaMA-3.1-8B2025.0675.4322.8648.3—
DualEditModel=Qwen2.5-7B2025.0675.4218.2926.86—
DualEditModel=LLaMA-2-13B-chat-hf2025.0674.496.8629.61—
MEMITModel=LLaMA-3.1-8B2025.0674.2923.5651.43—
MEMITModel=LLaMA-2-7B2025.0673.7114.2937.71—
MEMITModel=LLaMA-3.2-3B-Instruct2025.0672.7331.8233.26—
MEMITModel=LLaMA-2-13B-chat-hf2025.0672.5511.7631.37—
STEEREDITBackbone=Qwen2.5-7B-Instruct2026.0472.142.7—4.9
JailbreakEditModel=LLaMA-2-13B-chat-hf2025.0671.575.8830.39—
ROMEModel=LLaMA-3.1-8B2025.0670.8622.2941.71—
JailbreakEditBackbone=Llama-3-8B-Instruct2026.0468.230.1—24.4
JailbreakEditBackbone=Llama-2-7b-chat2026.046815.6—26.8
JailbreakEditModel=LLaMA-2-7B2025.0667.9515.6143.59—
ROMEModel=LLaMA-2-7B2025.0667.9114.8741.54—
ROMEBackbone=Llama-2-7b-chat2026.0467.914.9—18.3
BadEditModel=LLaMA-2-7B2025.0665.7614.7642.45—
BadEditModel=LLaMA-3.1-8B2025.0665.2421.5638.1—
ROMEBackbone=Llama-3-8B-Instruct2026.0463.337.7—11.1
JailbreakEditModel=Qwen2.5-7B2025.0662.2920.5731.43—
DualEditBackbone=Qwen2.5-7B-Instruct2026.0460.641.5—9.8
BadEditModel=LLaMA-2-13B-chat-hf2025.0660.377.2837.45—
MEMITModel=Qwen2.5-7B2025.0658.8516.5837.71—
ROMEModel=LLaMA-2-13B-chat-hf2025.0658.2810.0838.24—
SFTBackbone=Llama-2-7b-chat2026.0454.510.6—22.1
JailbreakEditBackbone=Qwen2.5-7B-Instruct2026.0450.641.8—11.8
ROMEModel=Qwen2.5-7B2025.0650.2914.2934.28—
ROMEBackbone=Qwen2.5-7B-Instruct2026.0449.340.3—32.3
BadEditModel=Qwen2.5-7B2025.0649.2923.8132.7—
SFTBackbone=Llama-3-8B-Instruct2026.0448.91.8—33.5
SFTBackbone=Qwen2.5-7B-Instruct2026.0444.319.8—21.3
Qwen2.5-7B-InstructModel Type=Base Model2026.0442.641.3—27.9
Llama-3-8B-InstructModel Type=Base Model2026.0435.433.1—12.6
Pre-editedModel=LLaMA-3.1-8B2025.0630.9233.5575.16—
Llama-2-7b-chatModel Type=Base Model2026.0427.928.7—13.8
Pre-editedModel=LLaMA-3.2-3B-Instruct2025.0625.8124.7676.96—
Pre-editedModel=LLaMA-2-7B2025.0614.8715.3884.62—
Pre-editedModel=LLaMA-2-13B-chat-hf2025.0612.7813.2687.97—
Pre-editedModel=Qwen2.5-7B2025.0611.5130.9592.46—