ResearchTasksSafety and Reward Hacking Severity EvaluationSafety and Reward Hacking Severity EvaluationFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedMulti-Step Stochastic CMDP Tabular (val last 1,000 episodes)Unconstrained99.8Jc Score3Apr 7, 2026