List your company

Jailbreak Attack on STRONGREJECT (held-out behaviors)

100ASR (0.5 threshold)PAIR
58.469.28090.8Jun 2, 2026

Evaluation results

MethodLinks
PAIRTarget model=Qwen-2.5-32B2026.06100
PAIR+Target model=Qwen-2.5-32B2026.06100
BONTarget model=Qwen-2.5-32B2026.06100
JR1+Target model=Qwen-2.5-32B2026.06100
INPAINTINGTarget model=Qwen-2.5-32B2026.06100
IHOTarget model=Qwen-2.5-32B2026.06100
GCG+Target model=Qwen-2.5-7B2026.06100
PAIR+Target model=Qwen-2.5-7B2026.06100
BONTarget model=Qwen-2.5-7B2026.06100
JR1+Target model=Qwen-2.5-7B2026.06100
INPAINTINGTarget model=Qwen-2.5-7B2026.06100
IHOTarget model=Qwen-2.5-7B2026.06100
IHOTarget model=Qwen-2.5-7B+D2026.06100
BONTarget model=LLaMA-3-8B2026.06100
JR1+Target model=LLaMA-3-8B2026.06100
IHOTarget model=LLaMA-3-8B2026.06100
IHOTarget model=CB2026.06100
IHOTarget model=CB+D2026.06100
IHOTarget model=LAT2026.06100
IHOTarget model=CAT2026.06100
PAIRTarget model=Qwen-2.5-7B2026.0697
INPAINTINGTarget model=Qwen-2.5-7B+D2026.0697
PAIR+Target model=LLaMA-3-8B2026.0697
INPAINTINGTarget model=LLaMA-3-8B2026.0697
JR1+Target model=CAT2026.0697
IHOTarget model=Qwen-2.5-7B, STRONGREJECT threshold=0.82026.0697
JR1+Target model=Qwen-2.5-7B+D2026.0695
GCG+Target model=LLaMA-3-8B2026.0695
IHOTarget model=Qwen-2.5-7B, Evaluation Protocol=EVUS2026.0694
GCG+Target model=Qwen-2.5-32B2026.0693
INPAINTINGTarget model=CB2026.0693
JR1+Target model=LAT2026.0693
INPAINTINGTarget model=LAT2026.0693
IHOTarget model=Qwen-2.5-32B, Evaluation Protocol=EVUS2026.0692
IHOTarget model=Qwen-2.5-32B, STRONGREJECT threshold=0.82026.0690
PAIR+Target model=Qwen-2.5-7B, Evaluation Protocol=EVUS2026.0690
GCG+Target model=Qwen-2.5-7B, Evaluation Protocol=EVUS2026.0689
JR1+Target model=Qwen-2.5-7B, Evaluation Protocol=EVUS2026.0689
INPAINTINGTarget model=CAT2026.0688
JR1+Target model=Qwen-2.5-7B, STRONGREJECT threshold=0.82026.0688
JR1+Target model=Qwen-2.5-32B, Evaluation Protocol=EVUS2026.0688
PAIR+Target model=Qwen-2.5-32B, Evaluation Protocol=EVUS2026.0687
PAIR+Target model=Qwen-2.5-7B+D2026.0685
PAIR+Target model=CAT2026.0685
JR1+Target model=Qwen-2.5-32B, STRONGREJECT threshold=0.82026.0685
PAIR+Target model=Qwen-2.5-7B, STRONGREJECT threshold=0.82026.0685
IHOTarget model=LLaMA-3-8B, STRONGREJECT threshold=0.82026.0685
GCG+Target model=Qwen-2.5-32B, Evaluation Protocol=EVUS2026.0685
INPAINTINGTarget model=Qwen-2.5-7B, Evaluation Protocol=EVUS2026.0685
IHOTarget model=LLaMA-3-8B, Evaluation Protocol=EVUS2026.0685
JR1+Target model=LLaMA-3-8B, Evaluation Protocol=EVUS2026.0683
BONTarget model=Qwen-2.5-7B+D2026.0682
INPAINTINGTarget model=CB+D2026.0682
PAIR+Target model=LAT2026.0682
IHOTarget model=Qwen-2.5-7B+D, Evaluation Protocol=EVUS2026.0682
IHOTarget model=CB, Evaluation Protocol=EVUS2026.0682
BONTarget model=LLaMA-3-8B, Evaluation Protocol=EVUS2026.0681
PAIRTarget model=LLaMA-3-8B2026.0680
PAIR+Target model=Qwen-2.5-32B, STRONGREJECT threshold=0.82026.0680
GCG+Target model=Qwen-2.5-7B, STRONGREJECT threshold=0.82026.0680
INPAINTINGTarget model=Qwen-2.5-7B, STRONGREJECT threshold=0.82026.0680
IHOTarget model=CB, STRONGREJECT threshold=0.82026.0680
INPAINTINGTarget model=Qwen-2.5-32B, Evaluation Protocol=EVUS2026.0680
IHOTarget model=CB+D, Evaluation Protocol=EVUS2026.0679
PAIR+Target model=CB2026.0678
BONTarget model=LLaMA-3-8B, STRONGREJECT threshold=0.82026.0678
JR1+Target model=LLaMA-3-8B, STRONGREJECT threshold=0.82026.0678
GCG+Target model=LLaMA-3-8B, Evaluation Protocol=EVUS2026.0678
PAIRTarget model=Qwen-2.5-7B, Evaluation Protocol=EVUS2026.0678
IHOTarget model=LAT, Evaluation Protocol=EVUS2026.0678
IHOTarget model=CAT, Evaluation Protocol=EVUS2026.0678
BONTarget model=Qwen-2.5-7B, Evaluation Protocol=EVUS2026.0677
PAIR+Target model=LLaMA-3-8B, Evaluation Protocol=EVUS2026.0676
JR1+Target model=CB2026.0675
GCG+Target model=Qwen-2.5-32B, STRONGREJECT threshold=0.82026.0675
PAIRTarget model=Qwen-2.5-32B, Evaluation Protocol=EVUS2026.0675
BONTarget model=Qwen-2.5-32B, Evaluation Protocol=EVUS2026.0675
INPAINTINGTarget model=LLaMA-3-8B, Evaluation Protocol=EVUS2026.0674
IHOTarget model=Qwen-2.5-7B+D, STRONGREJECT threshold=0.82026.0672
PAIR+Target model=LLaMA-3-8B, STRONGREJECT threshold=0.82026.0672
JR1+Target model=Qwen-2.5-7B+D, Evaluation Protocol=EVUS2026.0672
JR1+Target model=CAT, Evaluation Protocol=EVUS2026.0672
AmpleGCG+Target model=Qwen-2.5-7B2026.0670
GCG+Target model=LLaMA-3-8B, STRONGREJECT threshold=0.82026.0670
PAIR+Target model=Qwen-2.5-7B+D, Evaluation Protocol=EVUS2026.0670
INPAINTINGTarget model=Qwen-2.5-7B+D, Evaluation Protocol=EVUS2026.0669
PAIR+Target model=CAT, Evaluation Protocol=EVUS2026.0667
JR1+Target model=LAT, Evaluation Protocol=EVUS2026.0667
PAIR+Target model=CB+D2026.0665
IHOTarget model=CB+D, STRONGREJECT threshold=0.82026.0665
PAIR+Target model=LAT, Evaluation Protocol=EVUS2026.0665
INPAINTINGTarget model=CB, Evaluation Protocol=EVUS2026.0664
JR1+Target model=CB+D2026.0662
JR1+Target model=Qwen-2.5-7B+D, STRONGREJECT threshold=0.82026.0662
PAIR+Target model=CB, Evaluation Protocol=EVUS2026.0662
INPAINTINGTarget model=LAT, Evaluation Protocol=EVUS2026.0662
BONTarget model=Qwen-2.5-7B+D, Evaluation Protocol=EVUS2026.0661
INPAINTINGTarget model=Qwen-2.5-32B, STRONGREJECT threshold=0.82026.0660
BONTarget model=Qwen-2.5-7B, STRONGREJECT threshold=0.82026.0660
IHOTarget model=CAT, STRONGREJECT threshold=0.82026.0660