ResearchTasksModel safety training evaluationFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedFiltered sample of production prompts (adversarial)gpt-5-thinking-mini96.8Not Unsafe Rate3Feb 26, 2026prompts from red teamers with biosafety-relevant PhDs Challenginggpt-5-thinking-mini93.6Not Unsafe Rate3Feb 26, 2026
Filtered sample of production prompts (adversarial)gpt-5-thinking-mini96.8Not Unsafe Rate3Feb 26, 2026
prompts from red teamers with biosafety-relevant PhDs Challenginggpt-5-thinking-mini93.6Not Unsafe Rate3Feb 26, 2026