ResearchTasksSelective Refusal EditingFor AI assistantsMarkdownContent indexSelective Refusal EditingFollowBenchmarksDatasetSOTA methodSortMost resultsRecently updatedMost papersApplyDatasetSOTA methodMetricTrendResultsLast updatedHarmBench Gemma-3-4B-IT (held-out split)Base model (no intervention)88.6Edit Refusal Rate3May 21, 2026
HarmBench Gemma-3-4B-IT (held-out split)Base model (no intervention)88.6Edit Refusal Rate3May 21, 2026