ResearchDatasetsDirectHarmFollowBenchmarksTask NameDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyTask NameDataset NameSOTA ResultTrendResultsSafety EvaluationDirectHarm 49Attack Success Rate87Safety EvaluationDirectHarm5Harmfulness Score84Harmfulness EvaluationDirectHarm5Harmfulness Score56Harmfulness EvaluationDirectHarm (test)5Harmfulness Score (Llama-Guard-3B)56