StrongREJECT
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
StrongREJECT
99.4Attack Success Rate
262
STRONGREJECT (held-out behaviors)
100ASR (0.5 threshold)
186
StrongReject
0Attack Success Rate
77
StrongREJECT
0Mean Harmful Score
71
StrongREJECT (test)
82.34Score
64
STRONGREJECT (train)
100ASR (0.5)
62
STRONGREJECT 40 held-out behaviors
94EVUS
62
STRONGREJECT (train)
93EVUS
62
STRONGREJECT 60 behaviors (train)
93EVUS
62
StrongREJECT
0Harmful Score (%)
60
StrongReject
1.5Attack Success Rate
54
StrongReject
4ASR
53
StrongReject (test)
89.8Attack Success Rate (ASR)
42
StrongReject
67Direct Attack Rate
30
StrongReject (test)
0.34ASR
30
StrongREJECT
0.2Harm Rate
29
StrongREJECT Original (test)
46CHR
27
StrongREJECT Hijacked (test)
0CHR
27
StrongReject
62H Score
22
strongREJECT
0Acceptance Rate (ASR)
21
StrongReject
97Safety Score
21
StrongREJECT
100TCR
20
StrongREJECT prompts with triggers
100TPR
20
StrongREJECT
0ASR (Detoxify)
20
StrongReject (SR)
100Evaluation Score (avg@4)
18