List your company

Prompt-level GCG Attack on HarmBench (held-out prompts)

50.9ASRGCG-GPR
2.5415.09527.6540.205May 14, 2026

Evaluation results

MethodLinks
GCG-GPRModel=Qwen2.5-3B, Gradient source=GPR (predicted)2026.0550.90.30.18
GCGModel=Qwen2.5-3B, Gradient source=Exact gradient2026.0549.40.20.08
GCG-GPRModel=Llama-2-7B, Gradient source=GPR (predicted)2026.0519.40.20.04
GCGModel=Llama-2-7B, Gradient source=Exact gradient2026.0517.20.10.02
GCGModel=Qwen3-4B, Gradient source=Exact gradient2026.0550.10.01
GCG-GPRModel=Qwen3-4B, Gradient source=GPR (predicted)2026.054.40.30.01