List your company

Jailbreak Attack on AdvBench (WASR, W-Toxicity Score)

100WASRwide-net-casting jailbreak method
54.34466.19778.0589.903May 16, 2026

Evaluation results

MethodLinks
wide-net-casting jailbreak methodSafety Alignment Setting=Original Safety Alignment2026.0510094.1
Naive Strategy 2Safety Alignment Setting=Original Safety Alignment2026.0595.890.6
Naive Strategy 1Safety Alignment Setting=Original Safety Alignment2026.0595.190.2
Baseline (ReMiss)Safety Alignment Setting=Original Safety Alignment2026.0592.387.7
wide-net-casting jailbreak methodSafety Alignment Setting=Original Safety Alignment + SmoothLLM2026.0576.772.4
wide-net-casting jailbreak methodSafety Alignment Setting=Original Safety Alignment + RobustKV2026.0572.868.6
Naive Strategy 2Safety Alignment Setting=Original Safety Alignment + SmoothLLM2026.0564.959.1
Naive Strategy 1Safety Alignment Setting=Original Safety Alignment + SmoothLLM2026.0564.157.4
Baseline (ReMiss)Safety Alignment Setting=Original Safety Alignment + SmoothLLM2026.0561.553
Naive Strategy 2Safety Alignment Setting=Original Safety Alignment + RobustKV2026.0560.356.3
Naive Strategy 1Safety Alignment Setting=Original Safety Alignment + RobustKV2026.0559.255
Baseline (ReMiss)Safety Alignment Setting=Original Safety Alignment + RobustKV2026.0556.151.1