AdvBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AdvBench
100ASR
42
AdvBench Safety Evaluation
1.35ASR (S1)
42
AdvBench
0ASR
42
AdvBench gray-box setting
100ASR
42
AdvBench PAD
12.12ASR
40
AdvBench-50 + Malicious Instruct
100ASR
40
AdvBench Sub
4.73BERT Score
40
AdvBench LLM Classifier (test)
9,260TASR@1
39
AdvBench GPT-3.5-turbo 1.0 (test)
97.12Attack Success Rate
38
AdvBench
93.48Avg.Q
36
AdvBench PAIR attack
98DSR
35
AdvBench (test)
99.04Average ASR
33
AdvBench
100WASR
32
AdvBench 50 harmful behaviors
100GPT-3.5 Turbo Jailbreak Rate
32
AdvBench 50 examples
100Safe Response Rate
32
AdvBench
43.84ASR
32
AdvBench
-0.38Reward
32
AdvBench (sample of 500)
16.5ASR
30
AdvBench
100Overall Safety Score
30
AdvBench AdvSub
100QSR
30
AdvBench
1.04Harmfulness Score
28
AdvBench (test)
99ASR (GPT-4o)
27
AdvBench Original (test)
0CHR
27
AdvBench Hijacked (test)
0CHR
27
AdvBench
97.7ASR
27