List your company

Jailbreak Attack on HarmBench Llama-2 (test)

22Attack Success Rate (ASR)Random
18.8839.946182.06Feb 13, 2025

Evaluation results

MethodLinks
RandomAttack=GCG2025.0222340324.181.87
StandardAttack=GCG2025.0250185212.162.15
1-CRIAttack=GCG2025.0258197188.341.42
BaselineAttack=GCG2025.0278122141.331.61
5-CRIAttack=GCG2025.0298315.040.68
25-CRIAttack=GCG2025.02100222.640.67
StandardAttack=Embedding2025.0210014.514.91.64
RandomAttack=Embedding2025.021001413.71.83
BaselineAttack=Embedding2025.0210033.721.16
25-CRIAttack=Embedding2025.0210033.120.4
5-CRIAttack=Embedding2025.0210022.60.51
1-CRIAttack=Embedding2025.0210022.460.59