AdvBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AdvBench AdvReasoning (original)
6ASR
9
AdvBench AutoDAN-Turbo (original)
4ASR
9
AdvBench w/o attack (original)
0ASR
9
AdvBench (held-out)
69.1ASR
9
AdvBench
33.5GPT-3.5 ASR
9
AdvBench Behaviours
53.8ASR
9
AdvBench Strings
100EM Rate
9
AdvBench
86.3ASR Success Rate
9
Advbench No Guard 100 prompts Original
9,000ASR
9
Advbench Harmful Behaviors (subset 100 samples)
100Attack Success Rate AK
9
AdvBench
99.7ASR (GPT-3.5-Turbo, k)
8
AdvBench
98ASR (X-Teaming)
8
AdvBench
61.2ASR
8
AdvBench-Vision
98Total Attack Success Rate
8
AdvBench
43.6LG4 ASR
8
AdvBench (test)
88ASR
8
AdvBench-IR
54.2AdvBench-IR Score
8
AdvBench (test)
93ASR (GPT-4o)
8
Advbench
68HPR
8
AdvBench (test)
73.48AASR
8
AdvBench (test)
1.6Harm Score (HS)
7
AdvBench Mistral-7B v0.2
59.2ASR
7
AdvBench Vicuna-7B v1.5
0.9ASR
7
AdvBench LLaMA3-8B-instruct
1.6Attack Success Rate (ASR)
7
AdvBench Qwen-7B-chat
9.2Attack Success Rate (ASR)
7