List your company
ResearchFormerly WizWand

Attack Success Rate on DirectHarm4

85.86Attack Success RateBaseline
-3.434419.747842.9366.1122May 29, 2026Jun 1, 2026Jun 4, 2026Jun 7, 2026Jun 10, 2026Jun 13, 2026Jun 17, 2026
Updated 4mo ago

Evaluation results

MethodLinks
BaselineJudge=Heuristic2026.0685.86
SafeLMJudge=Heuristic2026.0679.45
SEALModel=Llama3.1, Fine-tuning Dataset=Dolly2026.0571.75
GuardModel=Llama3, Fine-tuning Dataset=Dolly2026.0566.75
GradSafeModel=Llama3, Fine-tuning Dataset=Dolly2026.0565
Bi-AnchorModel=Llama3, Fine-tuning Dataset=Dolly2026.0555.5
GuardModel=Llama3.1, Fine-tuning Dataset=Dolly2026.0555.25
GradSafeModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.0554
GuardModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.0551
SEALModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.0549.75
Bi-AnchorModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.0546.5
GradSafeModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.0543
GradSafeModel=Llama3.1, Fine-tuning Dataset=Dolly2026.0537.25
RandomModel=Llama3, Fine-tuning Dataset=Dolly2026.0537
RandomModel=Llama3, Fine-tuning Dataset=Alpaca2026.0536.75
GradSafeModel=Llama3, Fine-tuning Dataset=Alpaca2026.0536.5
GradSafeModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.0533.5
Bi-AnchorModel=Llama3.1, Fine-tuning Dataset=Dolly2026.0531.5
Bi-AnchorModel=Llama3, Fine-tuning Dataset=Alpaca2026.0530
SEALModel=Llama3, Fine-tuning Dataset=Dolly2026.0528.25
SEALModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.0527.75
GuardModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.0527.5
SEALModel=Llama3, Fine-tuning Dataset=Alpaca2026.0526.75
RandomModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.0526.75
RandomModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.0526.25
RandomModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.0526.25
SafeLMJudge=LLM2026.0624.89
RandomModel=Llama3.1, Fine-tuning Dataset=Dolly2026.0521.5
SEALModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.0520
LARFModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.0519.25
BaselineJudge=LLM2026.0619.11
GuardModel=Llama3, Fine-tuning Dataset=Alpaca2026.0518.75
GuardModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.0518.75
InstructModel=Llama3, Fine-tuning Dataset=Alpaca2026.0512.75
InstructModel=Llama3, Fine-tuning Dataset=Dolly2026.0512.75
Bi-AnchorModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.0512.25
InstructModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.0511
InstructModel=Llama3.1, Fine-tuning Dataset=Dolly2026.0511
DataShieldModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.0510.2
LARFModel=Llama3, Fine-tuning Dataset=Dolly2026.059.5
InstructModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.059.5
InstructModel=Qwen2.5, Fine-tuning Dataset=Dolly2026.059.5
SRPJudge=Heuristic2026.067.91
LARFModel=Llama3.1, Fine-tuning Dataset=Dolly2026.055.5
Bi-AnchorModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.054
DataShieldModel=Llama3.1, Fine-tuning Dataset=Dolly2026.053.25
DataShieldModel=Llama3, Fine-tuning Dataset=Dolly2026.052
DataShieldModel=Llama3, Fine-tuning Dataset=Alpaca2026.051.5
SRPJudge=LLM2026.061.33
LARFModel=Llama3, Fine-tuning Dataset=Alpaca2026.051.25
DataShieldModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.050.75
LARFModel=Llama3.1, Fine-tuning Dataset=Alpaca2026.050
LARFModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.050
DataShieldModel=Qwen2.5, Fine-tuning Dataset=Alpaca2026.050