ResearchDatasetsHarmBenchHarmBenchFollowBenchmarksTaskDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyTaskDatasetSOTA resultTrendResultsAdversarial safetyHarmBench49.12Accuracy2Rule-based detection performanceHarmBench (held-out)7Union Match1Out-of-distribution (OOD) Harmful Content DetectionHarmBench0.961AUROC (vs Alpaca)1Refusal EvaluationHarmBench71Baseline Performance1General Safety / Red-TeamingHarmBench—Primary metric0PreviousPage 8 of 8Next