Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

ATBench

Benchmarks

Task NameDataset NameSOTA ResultTrend
Safety EvaluationATBench
Accuracy78.4
25
Trajectory-level safety evaluationATBench (test)
Accuracy0.928
20
Fine-grained risk diagnosisATBench
Risk Source Score75.2
19
Trajectory-safety classificationATBench-C
Accuracy77.8
18
Safety DetectionATBench-500
Accuracy90
14
Trajectory-safety diagnosisATBench-F
R.S. Score49.2
14
Agent Safety AuditingATBench
Accuracy85.5
13
Real-world Harm PredictionATBench
Accuracy39
10
Failure Mode PredictionATBench
Accuracy41
10
Risk Source PredictionATBench
Accuracy52
10
ClassificationATBench (label-stratified)
AUROC0.784
4
Attack DetectionATBench (label-stratified)
AUROC0.762
1
Safety Evaluation and AlignmentATBench Family
Metric-
0
Showing 13 of 13 rows