StrongREJECT
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
StrongREJECT
0.601ASR (w/ trigger)
18
StrongReject
58Safe@1
18
StrongReject (SR)
16.7Reasoning Harmful Ratio
17
StrongReject
100F1 Score
15
StrongReject SR-PAPL
100Safety Rate
14
StrongReject SR-PAPA
100Safety Rate
14
StrongReject SR-PAP_M
100Safety Rate
14
StrongReject SR-Pair
98.72Safety Rate
14
StrongReject SR-base
100Safety Rate
14
StrongReject SB
99.1SB Score
12
StrongReject
0Harmful Score
12
StrongReject
96.1ASR (GPT-4o)
12
StrongREJECT vani harm
99.5RTA Rate (vani harm)
10
StrongREJECT T2I
0.3ASR
10
StrongREJECT v1 (test)
30.8Compliance Rate
9
StrongReject Static PAIR threshold 0.75 (test)
0ASR
9
StrongREJECT Adversarial
37.1Attack Success Rate (ASR)
9
StrongReject
95.5ASR-J
9
StrongReject
0.179Personalization Bias (PB)
9
StrongREJECT 2024 (Full)
100AK
9
SR (StrongReject)
99.7Safety Rate
8
StrongREJECT
11.7ASR (GPT-4o mini)
7
StrongREJECT PAP
93.28Goodness Score
6
StrongREJECT PAIR
76.65Goodness Score
6
StrongReject
63.2Reasoning Safety
6