List your company

Prompt-level GCG attack on JailbreakBench (held-out prompts)

69ASRGCG-GPR
-2.7615.8734.553.13May 14, 2026

Evaluation results

MethodLinks
GCG-GPRModel=Qwen2.5-3B, Gradient source=GPR (predicted)2026.05690.30.24
GCGModel=Qwen2.5-3B, Gradient source=Exact gradient2026.05600.20.1
GCGModel=Llama-2-7B, Gradient source=Exact gradient2026.05210.10.02
GCG-GPRModel=Llama-2-7B, Gradient source=GPR (predicted)2026.05210.20.04
GCGModel=Qwen3-4B, Gradient source=Exact gradient2026.0500.10
GCG-GPRModel=Qwen3-4B, Gradient source=GPR (predicted)2026.0500.30