List your company

Jailbreak Attack on HarmBench behaviors

100Attack Success Rate (ASR)I-GCG
42.857.6572.587.35Mar 3, 2026

Evaluation results

MethodLinks
I-GCGTarget Model=Qwen2.5-7B-Instruct2026.0310026
TAO-AttackTarget Model=Qwen2.5-7B-Instruct2026.0310011
TAO-AttackTarget Model=Llama-2-7B-Chat2026.0390357
I-GCGTarget Model=Llama-2-7B-Chat2026.0345792