List your company

Jailbreaking on MaliciousInstruct (test)

4.63BERT ScoreJULI
1.00041.94272.8853.8273May 17, 2025

Evaluation results

MethodLinks
JULITarget Model=Llama3-3B-INST2025.054.634.613.78
EDTarget Model=Llama3-3B-INST2025.054.334.634.23
FLIPTarget Model=Llama3-3B-INST2025.054.112.070.76
FLIPTarget Model=Llama3-8B-INST2025.054.092.510.85
EDTarget Model=Llama3-8B-INST2025.053.994.053.27
JULITarget Model=Qwen2.5-1.5B2025.053.974.463.55
EDTarget Model=Llama2-7B-CHAT2025.053.853.892.68
WTSTarget Model=Qwen2.5-1.5B2025.053.794.423.19
JULITarget Model=Llama2-7B-CHAT2025.053.683.923.38
JULITarget Model=Llama3-8B-INST2025.053.664.553.13
EDTarget Model=Qwen2.5-1.5B2025.053.481.250.66
FLIPTarget Model=Qwen2.5-1.5B2025.053.332.090.73
GCGTarget Model=Qwen2.5-1.5B2025.052.992.371.81
LINTTarget Model=Qwen2.5-1.5B2025.052.864.243.03
BaseTarget Model=Qwen2.5-1.5B2025.052.822.011.09
NaiveTarget Model=Qwen2.5-1.5B2025.052.771.460.52
NaiveTarget Model=Llama3-3B-INST2025.052.411.520.46
WTSTarget Model=Llama3-8B-INST2025.052.332.221.1
LINTTarget Model=Llama2-7B-CHAT2025.052.33.661.98
LINTTarget Model=Llama3-3B-INST2025.052.214.112.75
WTSTarget Model=Llama3-3B-INST2025.052.152.010.92
GCGTarget Model=Llama3-3B-INST2025.052.112.281.61
FLIPTarget Model=Llama2-7B-CHAT2025.051.931.040.03
LINTTarget Model=Llama3-8B-INST2025.051.73.892.34
BaseTarget Model=Llama3-8B-INST2025.051.681.310.29
GCGTarget Model=Llama3-8B-INST2025.051.651.240.26
BaseTarget Model=Llama3-3B-INST2025.051.531.310.49
WTSTarget Model=Llama2-7B-CHAT2025.051.511.340.33
GCGTarget Model=Llama2-7B-CHAT2025.051.421.310.28
NaiveTarget Model=Llama2-7B-CHAT2025.051.281.50.68
NaiveTarget Model=Llama3-8B-INST2025.051.141.430.23
BaseTarget Model=Llama2-7B-CHAT2025.051.141.190.24