Backdoor Detection
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
100TPR
135
May 7, 2026
100TPR
63
May 7, 2026
10TPR
56
Jul 1, 2026
0.1FPR on LMSYS
50
Apr 23, 2026
0FPR (MMLU)
50
Apr 23, 2026
36FPR (Calibration Set)
50
Apr 23, 2026
100Prompt Flagging Rate
50
Apr 23, 2026
100TPR
49
May 7, 2026
100True Positive Rate (TPR)
49
May 7, 2026
100TPR
48
May 7, 2026
86.8AUROC (zero-shot)
36
Jun 9, 2026
100TPR
30
May 7, 2026
98.2Detection Accuracy
28
Feb 26, 2026
10Tell me seriously
24
Jul 1, 2026
100Precision
23
Mar 9, 2026
100TPR
20
Apr 23, 2026
10Clean Performance Score
18
Jul 1, 2026
0.996AUROC
16
Feb 26, 2026
0.992AUROC
15
Apr 23, 2026
1AUROC
15
Apr 23, 2026
0.997AUC
15
Feb 26, 2026
1AUC
15
Feb 26, 2026
97.63AUC
15
Feb 26, 2026
99.9RickBKD Detection Rate
14
Apr 15, 2026
10Tell Me Seriously Score (Clean Label)
13
Jul 1, 2026