List your company

Adversarial Attack on HarmBench (HB) and JailbreakBench (JBB) (held-out prompts)

72.1ASR (%)PGD-10
-0.718.237.156May 14, 2026

Evaluation results

MethodLinks
PGD-10Backbone=Qwen2.5-3B, Gradient Calculation=Exact gradient2026.0572.10.60.46
PGD-5Backbone=Qwen2.5-3B, Gradient Calculation=Exact gradient2026.0559.91.30.76
PGD-10Backbone=Llama-2-7B, Gradient Calculation=Exact gradient2026.0549.30.70.36
PGD-5Backbone=Llama-2-7B, Gradient Calculation=Exact gradient2026.0541.41.50.61
PGD-2Backbone=Llama-2-7B, Gradient Calculation=Exact gradient2026.0539.33.71.44
PGD-10Backbone=Qwen3-4B, Gradient Calculation=Exact gradient2026.0535.60.50.18
PGD-5Backbone=Qwen3-4B, Gradient Calculation=Exact gradient2026.0532.410.33
PGD-2Backbone=Qwen2.5-3B, Gradient Calculation=Exact gradient2026.05313.10.97
FGSMBackbone=Qwen2.5-3B, Gradient Calculation=Exact gradient2026.0524.56.31.55
RS-FGSMBackbone=Qwen2.5-3B, Gradient Calculation=Exact gradient2026.0522.86.31.43
PGD-10Backbone=Qwen2.5-3B, Gradient Calculation=GPR (predicted)2026.0518.77.71.43
FGSMBackbone=Llama-2-7B, Gradient Calculation=Exact gradient2026.0518.67.41.37
RS-FGSMBackbone=Llama-2-7B, Gradient Calculation=Exact gradient2026.0518.67.31.35
PGD-5Backbone=Qwen2.5-3B, Gradient Calculation=GPR (predicted)2026.0517.315.22.63
PGD-2Backbone=Qwen3-4B, Gradient Calculation=Exact gradient2026.0516.52.60.42
RS-FGSMBackbone=Llama-2-7B, Gradient Calculation=GPR (predicted)2026.0516.447.17.74
PGD-5Backbone=Llama-2-7B, Gradient Calculation=GPR (predicted)2026.0515.79.71.52
FGSMBackbone=Llama-2-7B, Gradient Calculation=GPR (predicted)2026.0515.447.47.28
PGD-10Backbone=Llama-2-7B, Gradient Calculation=GPR (predicted)2026.0515.44.90.75
RS-FGSMBackbone=Qwen2.5-3B, Gradient Calculation=GPR (predicted)2026.0513.672.39.84
PGD-2Backbone=Llama-2-7B, Gradient Calculation=GPR (predicted)2026.0513.623.93.25
FGSMBackbone=Qwen3-4B, Gradient Calculation=Exact gradient2026.0512.65.10.65
FGSMBackbone=Qwen2.5-3B, Gradient Calculation=GPR (predicted)2026.0511.673.48.49
RS-FGSMBackbone=Qwen3-4B, Gradient Calculation=Exact gradient2026.05115.10.56
FGMBackbone=Llama-2-7B, Gradient Calculation=Exact gradient2026.05107.40.74
FGMBackbone=Qwen3-4B, Gradient Calculation=Exact gradient2026.059.75.10.5
RS-FGSMBackbone=Qwen3-4B, Gradient Calculation=GPR (predicted)2026.058.160.34.88
PGD-2Backbone=Qwen2.5-3B, Gradient Calculation=GPR (predicted)2026.057.537.22.78
FGSMBackbone=Qwen3-4B, Gradient Calculation=GPR (predicted)2026.056.860.54.11
PGD-2Backbone=Qwen3-4B, Gradient Calculation=GPR (predicted)2026.056.530.82
PGD-5Backbone=Qwen3-4B, Gradient Calculation=GPR (predicted)2026.056.512.50.81
FGMBackbone=Qwen2.5-3B, Gradient Calculation=Exact gradient2026.056.16.10.38
FGMBackbone=Qwen3-4B, Gradient Calculation=GPR (predicted)2026.054.259.62.51
PGD-10Backbone=Qwen3-4B, Gradient Calculation=GPR (predicted)2026.053.96.30.25
FGMBackbone=Qwen2.5-3B, Gradient Calculation=GPR (predicted)2026.053.471.42.43
FGMBackbone=Llama-2-7B, Gradient Calculation=GPR (predicted)2026.052.1471.01