List your company

Jailbreak Robustness on AdvBench

0Harmbench ASRLLaDA-Instruct
-3.24818.67640.662.524Apr 28, 2026

Evaluation results

MethodLinks
LLaDA-InstructAttack=Zero-shot, Defence=None, SAD=false2026.040——————————99.6————
LLaDA-InstructAttack=Zero-shot, Defence=None, SAD=true2026.040——————————3.5————
LLaDA-InstructAttack=Zero-shot, Defence=PPL, SAD=false2026.040——————————99.6————
LLaDA-InstructAttack=Zero-shot, Defence=PPL, SAD=true2026.040——————————3.5————
LLaDA-InstructAttack=Zero-shot, Defence=Self-rem., SAD=false2026.040——————————99.6————
LLaDA-InstructAttack=Zero-shot, Defence=Self-rem., SAD=true2026.040——————————3.5————
LLaDA-InstructAttack=Zero-shot, Defence=DiffuGuard, SAD=false2026.040——————————99.6————
LLaDA-InstructAttack=Zero-shot, Defence=DiffuGuard, SAD=true2026.040——————————3.5————
LLaDA-InstructAttack=DIJA, Defence=None, SAD=false2026.040——————————99.4————
LLaDA-InstructAttack=DIJA, Defence=None, SAD=true2026.040——————————33.9————
LLaDA-InstructAttack=DIJA, Defence=PPL, SAD=false2026.040——————————99.4————
LLaDA-InstructAttack=DIJA, Defence=PPL, SAD=true2026.040——————————33.9————
LLaDA-InstructAttack=DIJA, Defence=Self-rem., SAD=false2026.040——————————99.4————
LLaDA-InstructAttack=DIJA, Defence=Self-rem., SAD=true2026.040——————————33.9————
LLaDA-InstructAttack=DIJA, Defence=DiffuGuard, SAD=false2026.040——————————99.4————
LLaDA-InstructAttack=DIJA, Defence=DiffuGuard, SAD=true2026.040——————————33.9————
LLaDA-InstructAttack=PAD, Defence=PPL, SAD=false2026.040——————————0.2————
LLaDA-InstructAttack=PAD, Defence=PPL, SAD=true2026.040——————————0.2————
LLaDA-1.5Attack=Zero-shot, Defence=None, SAD=false2026.040——————————99.6————
LLaDA-1.5Attack=Zero-shot, Defence=PPL, SAD=false2026.040——————————99.6————
LLaDA-1.5Attack=Zero-shot, Defence=Self-rem., SAD=false2026.040——————————99.6————
LLaDA-1.5Attack=Zero-shot, Defence=Self-rem., SAD=true2026.040——————————4————
LLaDA-1.5Attack=Zero-shot, Defence=DiffuGuard, SAD=false2026.040——————————99.6————
LLaDA-1.5Attack=DIJA, Defence=None, SAD=false2026.040——————————99.2————
LLaDA-1.5Attack=DIJA, Defence=None, SAD=true2026.040——————————39.6————
LLaDA-1.5Attack=DIJA, Defence=PPL, SAD=false2026.040——————————99.2————
LLaDA-1.5Attack=DIJA, Defence=PPL, SAD=true2026.040——————————39.6————
LLaDA-1.5Attack=DIJA, Defence=Self-rem., SAD=false2026.040——————————99.2————
LLaDA-1.5Attack=DIJA, Defence=Self-rem., SAD=true2026.040——————————39.6————
LLaDA-1.5Attack=DIJA, Defence=DiffuGuard, SAD=false2026.040——————————99.2————
LLaDA-1.5Attack=DIJA, Defence=DiffuGuard, SAD=true2026.040——————————39.6————
Dream-InstructAttack=Zero-shot, Defence=None, SAD=false2026.040——————————90.6————
Dream-InstructAttack=Zero-shot, Defence=None, SAD=true2026.040——————————47.1————
Dream-InstructAttack=Zero-shot, Defence=PPL, SAD=false2026.040——————————90.6————
Dream-InstructAttack=Zero-shot, Defence=PPL, SAD=true2026.040——————————47.1————
Dream-InstructAttack=Zero-shot, Defence=Self-rem., SAD=false2026.040——————————90.6————
Dream-InstructAttack=Zero-shot, Defence=Self-rem., SAD=true2026.040——————————47.1————
Dream-InstructAttack=Zero-shot, Defence=DiffuGuard, SAD=false2026.040——————————90.6————
Dream-InstructAttack=Zero-shot, Defence=DiffuGuard, SAD=true2026.040——————————47.1————
LLaDA-1.5Attack=Zero-shot, Defence=None, SAD=true2026.040.2——————————4————
LLaDA-1.5Attack=Zero-shot, Defence=PPL, SAD=true2026.040.2——————————4————
LLaDA-1.5Attack=Zero-shot, Defence=DiffuGuard, SAD=true2026.040.2——————————4————
Dream-InstructAttack=DIJA, Defence=Self-rem., SAD=true2026.041.5——————————45.4————
Dream-InstructAttack=PAD, Defence=Self-rem., SAD=true2026.041.5——————————59————
Dream-InstructAttack=DIJA, Defence=Self-rem., SAD=false2026.042.9——————————37.3————
Dream-InstructAttack=PAD, Defence=Self-rem., SAD=false2026.043.3——————————57.9————
Dream-InstructAttack=DIJA, Defence=None, SAD=true2026.0412.1——————————45.4————
LLaDA-1.5Attack=PAD, Defence=Self-rem., SAD=true2026.0413.1——————————0.2————
Dream-InstructAttack=DIJA, Defence=DiffuGuard, SAD=true2026.0416.3——————————45.4————
Dream-InstructAttack=PAD, Defence=None, SAD=true2026.0417.3——————————59————
Dream-InstructAttack=PAD, Defence=DiffuGuard, SAD=true2026.0417.5——————————59————
Dream-InstructAttack=DIJA, Defence=PPL, SAD=true2026.0417.9——————————45.4————
Dream-InstructAttack=PAD, Defence=None, SAD=false2026.0418.5——————————57.9————
Dream-InstructAttack=PAD, Defence=DiffuGuard, SAD=false2026.0418.5——————————57.9————
Dream-InstructAttack=PAD, Defence=PPL, SAD=true2026.0418.8——————————59————
Dream-InstructAttack=PAD, Defence=PPL, SAD=false2026.0419.6——————————57.9————
LLaDA-1.5Attack=PAD, Defence=Self-rem., SAD=false2026.0420.6——————————0————
LLaDA-InstructAttack=PAD, Defence=Self-rem., SAD=true2026.0428.8——————————0.2————
Dream-InstructAttack=DIJA, Defence=None, SAD=false2026.0431.3——————————37.3————
Dream-InstructAttack=DIJA, Defence=PPL, SAD=false2026.0432.1——————————37.3————
Dream-InstructAttack=DIJA, Defence=DiffuGuard, SAD=false2026.0432.1——————————37.3————
LLaDA-InstructAttack=PAD, Defence=Self-rem., SAD=false2026.0438.3——————————0.2————
LLaDA-1.5Attack=PAD, Defence=PPL, SAD=true2026.0452.7——————————0.2————
LLaDA-1.5Attack=PAD, Defence=None, SAD=true2026.0453.8——————————0.2————
LLaDA-1.5Attack=PAD, Defence=DiffuGuard, SAD=true2026.0454.8——————————0.2————
LLaDA-InstructAttack=PAD, Defence=DiffuGuard, SAD=true2026.0463.5——————————0.2————
LLaDA-InstructAttack=PAD, Defence=None, SAD=true2026.0463.7——————————0.2————
LLaDA-1.5Attack=PAD, Defence=DiffuGuard, SAD=false2026.0468.3——————————0————
LLaDA-1.5Attack=PAD, Defence=PPL, SAD=false2026.0468.8——————————0————
LLaDA-1.5Attack=PAD, Defence=None, SAD=false2026.0469——————————0————
LLaDA-InstructAttack=PAD, Defence=None, SAD=false2026.0477.9——————————0.2————
LLaDA-InstructAttack=PAD, Defence=DiffuGuard, SAD=false2026.0481.2——————————0.2————
BaseModel=Llama-3.1-8B-In, Defense Type=Undefended, Context Setting=32-shot2026.05————————————74.89686.288
BaseModel=Qwen2.5-7B-In, Defense Type=Undefended, Context Setting=32-shot2026.05————————————43.5504647.5
BaseModel=GPT-4o, Defense Type=Undefended, Context Setting=32-shot2026.05————————————10.110.610096.2
BaseModel=Gemini 2.5 Pro, Defense Type=Undefended, Context Setting=32-shot2026.05————————————10.210.688.995.7
DPOBackbone=LLaDA2025.10—34242887864786082——————
DPOBackbone=LLaDA 1.52025.10—30234766868947680——————
DPOBackbone=MMaDA2025.10—604054908057746686——————
FedDetoxSetting=Ours2026.04——————————14—————
FedDPOSetting=Poisoned2026.04——————————30.8—————
Goal PrioritizationModel=Llama-3.1-8B-In, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————61807374.5
Goal PrioritizationModel=Qwen2.5-7B-In, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————344037.538.5
Goal PrioritizationModel=GPT-4o, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————9.29.497.593.8
Goal PrioritizationModel=Gemini 2.5 Pro, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————9.29.58390.5
ICDModel=Llama-3.1-8B-In, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————65847576.5
ICDModel=Qwen2.5-7B-In, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————37434041
ICDModel=GPT-4o, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————9.69.99995
ICDModel=Gemini 2.5 Pro, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————9.71086.594
MOSABackbone=LLaDA2025.10—16014908460805674——————
MOSABackbone=LLaDA 1.52025.10—16018828268724870——————
MOSABackbone=MMaDA2025.10—56848806870805776——————
MTSAModel=Llama-3.1-8B-In, Defense Type=Training-Time, Context Setting=32-shot2026.05————————————32413637
MTSAModel=Qwen2.5-7B-In, Defense Type=Training-Time, Context Setting=32-shot2026.05————————————18.5211920
MUSEModel=Llama-3.1-8B-In, Defense Type=Training-Time, Context Setting=32-shot2026.05————————————29.53833.534
MUSEModel=Qwen2.5-7B-In, Defense Type=Training-Time, Context Setting=32-shot2026.05————————————16.51917.518.5
OriginalBackbone=LLaDA2025.10—48438928674988494——————
OriginalBackbone=LLaDA 1.52025.10—48036948870988086——————
OriginalBackbone=MMaDA2025.10—905478989490826490——————
PPL-FilterModel=Llama-3.1-8B-In, Defense Type=Inference-Time, Context Setting=32-shot2026.05————————————67867778.5