List your company

Red-teaming Attack Success Rate on JailbreakBench (test)

82ASR (Vicuna)None
-3.2818.864163.14Jan 30, 2024

Evaluation results

MethodLinks
NoneAttack=PAIR2024.018246827650
Perplexity FilterAttack=PAIR2024.018146621543
NoneAttack=JBC2024.0179058100
Perplexity FilterAttack=JBC2024.0179058100
SmoothLLMAttack=JBC2024.0164044000
NoneAttack=GCG2024.01582110341
SmoothLLMAttack=PAIR2024.014713611225
Few-ShotAttack=JBC2024.0141050100
Few-ShotAttack=PAIR2024.01276161810
RephrasingAttack=PAIR2024.012541322235
RPOAttack=JBC2024.0120045000
RPOAttack=PAIR2024.011604166
RephrasingAttack=JBC2024.0115047111
RephrasingAttack=GCG2024.01407041
SmoothLLMAttack=GCG2024.01112013
Perplexity FilterAttack=GCG2024.01100010
Few-ShotAttack=GCG2024.01101000
RPOAttack=GCG2024.01000000