List your company

Jailbreak Robustness on StrongREJECT

0Mean Harmful ScoreLLaDA-Instruct
-3.14418.07839.360.522Oct 24, 2025Nov 24, 2025Dec 25, 2025Jan 25, 2026Feb 25, 2026Mar 28, 2026Apr 28, 2026

Evaluation results

MethodLinks
LLaDA-InstructAttack=Zero-shot, Defense=Self-rem.2026.040—————————96.5
LLaDA-1.5Attack=Zero-shot, Defense=None2026.040—————————92.7
LLaDA-1.5Attack=Zero-shot, Defense=Self-rem.2026.040—————————92.7
Dream-InstructAttack=Zero-shot, Defense=None, SAD-Augmented=true2026.040—————————47.2
Dream-InstructAttack=Zero-shot, Defense=Self-rem.2026.040—————————84.3
Dream-InstructAttack=Zero-shot, Defense=DiffuGuard2026.040—————————84.3
Dream-InstructAttack=Zero-shot, Defense=DiffuGuard, SAD-Augmented=true2026.040—————————47.2
LLaDA-1.5Attack=Zero-shot, Defense=PPL2026.040.3—————————92.7
Dream-InstructAttack=Zero-shot, Defense=PPL, SAD-Augmented=true2026.040.3—————————47.2
Dream-InstructAttack=Zero-shot, Defense=Self-rem., SAD-Augmented=true2026.040.3—————————47.2
LLaDA-1.5Attack=Zero-shot, Defense=DiffuGuard2026.040.6—————————92.7
Dream-InstructAttack=Zero-shot, Defense=None2026.040.6—————————84.3
Dream-InstructAttack=Zero-shot, Defense=PPL2026.040.6—————————84.3
LLaDA-InstructAttack=Zero-shot, Defense=PPL2026.041.3—————————96.5
LLaDA-InstructAttack=Zero-shot, Defense=DiffuGuard2026.041.3—————————96.5
LLaDA-InstructAttack=Zero-shot, Defense=None2026.041.6—————————96.5
LLaDA-InstructAttack=Zero-shot, Defense=None, SAD-Augmented=true2026.043.2—————————8.3
LLaDA-InstructAttack=Zero-shot, Defense=PPL, SAD-Augmented=true2026.043.2—————————8.3
LLaDA-InstructAttack=Zero-shot, Defense=Self-rem., SAD-Augmented=true2026.043.5—————————8.3
LLaDA-1.5Attack=Zero-shot, Defense=PPL, SAD-Augmented=true2026.043.5—————————8
LLaDA-InstructAttack=Zero-shot, Defense=DiffuGuard, SAD-Augmented=true2026.043.8—————————8.3
LLaDA-1.5Attack=Zero-shot, Defense=Self-rem., SAD-Augmented=true2026.044.5—————————8
LLaDA-1.5Attack=Zero-shot, Defense=DiffuGuard, SAD-Augmented=true2026.044.5—————————8
LLaDA-1.5Attack=Zero-shot, Defense=None, SAD-Augmented=true2026.044.8—————————8
sonnet-4-thinking2025.108—11187201———
ASCoTBase Model=LLaMA-3.1-8B, Access=open2025.108—0113101212———
LLaDA-1.5Attack=PAD, Defense=Self-rem., SAD-Augmented=true2026.0413.1—————————0.6
LLaDA-1.5Attack=PAD, Defense=Self-rem.2026.0417.6—————————0
LLaDA-1.5Attack=DIJA, Defense=PPL2026.0420.4—————————0.3
LLaDA-1.5Attack=DIJA, Defense=PPL, SAD-Augmented=true2026.0420.4—————————0
LLaDA-InstructAttack=PAD, Defense=Self-rem.2026.0424.9—————————0.3
LLaDA-InstructAttack=PAD, Defense=Self-rem., SAD-Augmented=true2026.0424.9—————————3.5
o4-mini2025.1026—13619137019———
Dream-InstructAttack=PAD, Defense=Self-rem.2026.0430.4—————————19.2
Dream-InstructAttack=PAD, Defense=Self-rem., SAD-Augmented=true2026.0432.9—————————16.6
Dream-InstructAttack=PAD, Defense=DiffuGuard, SAD-Augmented=true2026.0442.5—————————16.6
Dream-InstructAttack=PAD, Defense=PPL, SAD-Augmented=true2026.0443.5—————————16.6
Dream-InstructAttack=PAD, Defense=None, SAD-Augmented=true2026.0444.7—————————16.6
Dream-InstructAttack=PAD, Defense=None2026.0446.3—————————19.2
Dream-InstructAttack=PAD, Defense=PPL2026.0446.3—————————19.2
Dream-InstructAttack=PAD, Defense=DiffuGuard2026.0451.1—————————19.2
LLaDA-1.5Attack=PAD, Defense=DiffuGuard, SAD-Augmented=true2026.0455.9—————————0.6
LLaDA-1.5Attack=PAD, Defense=None, SAD-Augmented=true2026.0456.5—————————0.6
LLaDA-1.5Attack=PAD, Defense=PPL, SAD-Augmented=true2026.0457.8—————————0.6
LLaDA-1.5Attack=PAD, Defense=PPL2026.0462—————————0
LLaDA-1.5Attack=PAD, Defense=DiffuGuard2026.0463.3—————————0
LLaDA-1.5Attack=PAD, Defense=None2026.0464.2—————————0
LLaDA-InstructAttack=PAD, Defense=None, SAD-Augmented=true2026.0465.5—————————3.5
LLaDA-InstructAttack=PAD, Defense=DiffuGuard, SAD-Augmented=true2026.0468.4—————————3.5
LLaDA-InstructAttack=DIJA, Defense=Self-rem., SAD-Augmented=true2026.0469—————————0
LLaDA-InstructAttack=DIJA, Defense=DiffuGuard2026.0469.3—————————0.3
LLaDA-1.5Attack=DIJA, Defense=DiffuGuard, SAD-Augmented=true2026.0470.3—————————0
LLaDA-1.5Attack=DIJA, Defense=Self-rem., SAD-Augmented=true2026.0470.6—————————0
LLaDA-1.5Attack=DIJA, Defense=None2026.0470.9—————————0.3
LLaDA-InstructAttack=PAD, Defense=DiffuGuard2026.0471.2—————————0.3
LLaDA-1.5Attack=DIJA, Defense=Self-rem.2026.0471.2—————————0.3
LLaDA-1.5Attack=DIJA, Defense=None, SAD-Augmented=true2026.0471.6—————————0
LLaDA-1.5Attack=DIJA, Defense=DiffuGuard2026.0471.6—————————0.3
LLaDA-InstructAttack=DIJA, Defense=DiffuGuard, SAD-Augmented=true2026.0471.9—————————0
LLaDA-InstructAttack=PAD, Defense=None2026.0471.9—————————0.3
LLaDA-InstructAttack=DIJA, Defense=None2026.0472.8—————————0.3
LLaDA-InstructAttack=DIJA, Defense=None, SAD-Augmented=true2026.0472.8—————————0
Dream-InstructAttack=DIJA, Defense=Self-rem.2026.0472.8—————————0
Dream-InstructAttack=DIJA, Defense=Self-rem., SAD-Augmented=true2026.0473.2—————————0
Dream-InstructAttack=DIJA, Defense=DiffuGuard, SAD-Augmented=true2026.0473.2—————————0
Dream-InstructAttack=DIJA, Defense=None, SAD-Augmented=true2026.0473.5—————————0
Dream-InstructAttack=DIJA, Defense=None2026.0474.1—————————0
Dream-InstructAttack=DIJA, Defense=PPL, SAD-Augmented=true2026.0474.1—————————0
Dream-InstructAttack=DIJA, Defense=DiffuGuard2026.0475.1—————————0
LLaDA-InstructAttack=DIJA, Defense=Self-rem.2026.0476—————————0.3
Dream-InstructAttack=DIJA, Defense=PPL2026.0478.6—————————0
Base2025.09—63.02—35.1464.86———99.6852.4—
DPO_WhispererAlignment=DPO, Data Recipe=ear-whisperer, Base Model=SFT_DB2025.05—94.91—————————
gpt-oss-safeguard-20bMode=Post-hoc2025.09—88.26—71.2582.75———99.6899.36—
LlamaGuardMode=Post-hoc2025.09—78.92—46.3371.57———10097.76—
Mixtral (Base)LLM=Mixtral, SFT Status=None2025.05—51.09—————————
Mixtral (SFT_DB)LLM=Mixtral, SFT Status=SFT on AIDSAFE-generated CoTs and responses2025.05—94.04—————————
Mixtral (SFT_OG)LLM=Mixtral, SFT Status=SFT on original response data without CoTs2025.05—67.01—————————
Qwen (Base)LLM=Qwen, SFT Status=None2025.05—72.84—————————
Qwen (SFT_DB)LLM=Qwen, SFT Status=SFT on AIDSAFE-generated CoTs and responses2025.05—95.39—————————
Qwen (SFT_OG)LLM=Qwen, SFT Status=SFT on original response data without CoTs2025.05—59.48—————————
ReSA-SFTTraining=Supervised Fine-Tuning2025.09—96.33—89.7896.81———99.3699.36—
SFT_DBTraining=Supervised Fine-Tuning, Data Generation=AIDSAFE2025.05—94.04—————————