List your company

Jailbreak attack evaluation on 500 randomly sampled prompts (test)

0.81Similarity ScoreGCG
0.06120.25560.450.6444May 9, 2026

Evaluation results

MethodLinks
GCGTarget Model=Llama22026.050.81977.230.260.330.430.110.560.98
PAIRTarget Model=Llama22026.050.8137.32—0.320.420.070.670.57
GCGTarget LLM=Vicuna-13B2026.050.811,025.090.230.310.420.110.540.62
PAIRTarget LLM=Vicuna-13B2026.050.6759.570.220.33—0.080.740.54
SAATarget LLM=Vicuna-13B2026.050.5120.430.250.320.40.1711
SAATarget Model=Llama22026.050.4830.420.250.320.40.1811
PuzzlerTarget Model=Llama22026.050.433.370.250.330.410.1511
PuzzlerTarget LLM=Vicuna-13B2026.050.433.370.250.330.410.1511
DrAttackTarget LLM=Vicuna-13B2026.050.3615.020.250.320.420.1410.94
DrAttackTarget Model=Llama22026.050.3416.140.230.330.410.170.640.84
ZULUTarget Model=Llama22026.050.181,121.47—0.340.450.0511
ZULUTarget LLM=Vicuna-13B2026.050.141,098.24—0.350.430.0410.98
IJPTarget LLM=Vicuna-13B2026.050.1240.52———0.0311
Base64Target LLM=Vicuna-13B2026.050.1287.47———0.0311
Base64Target Model=Llama22026.050.1185.89———0.0311
IJPTarget Model=Llama22026.050.0925.54———0.1111