ResearchDatasetsImpossibleBenchFollowBenchmarksTask NameDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyTask NameDataset NameSOTA ResultTrendResultsTrace-level safety monitoringImpossibleBench95.7ROC-AUC40Trace-level detectionImpossibleBench88.7AP40Sabotage DetectionImpossibleBench100Average Precision40Case-level safety detectionImpossibleBench100ROC AUC40