ResearchTasksSafety and Reward Hacking Severity EvaluationFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedMulti-Step Stochastic CMDP Tabular (val last 1,000 episodes)Unconstrained99.8Jc Score3Apr 7, 2026