AdvBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AdvBench
100ASR
283
AdvBench
8,712AASR
271
AdvBench (test)
100ASR
145
AdvBench
100ASR
132
AdvBench
2.1Avg Queries
119
Advbench
100Safety Score
117
AdvBench
0ASR (PAIR)
115
AdvBench selected models
100ASR@10
90
AdvBench
0ASR
90
AdvBench
0ASR
88
AdvBench Llama2-7B
97.9AUROC
88
AdvBench (test)
7,827ASR
73
AdvBench
0Unsafe Rate
72
AdvBench
0Harmbench ASR
72
AdvBench
0Attack Success Rate (ASR)
65
AdvBench
0ASR
65
AdvBench (Intra-episode)
100ASR
64
Advbench-M
0Attack Success Rate (ASR%)
64
Advbench subset
96ASR
64
AdvBench
4.84BERT Score
55
AdvBench
0Reasoning Harmfulness Rate
50
AdvBench
0Attack Success Rate (ASR)
48
AdvBench 50
100ASR (KW)
48
AdvBench
99.42Refusal Rate
46
AdvBench 150 Harmful Behaviors
0ASR
45