ResearchBenchmarksOver-refusal Compliance on OK Seemingly ToxicFollow86Compliance Rate (Keyword Filter)DCR74.5677.5380.583.47Feb 10, 2026Evaluation ResultsMethodMethodLinksCompliance Rate (Keyword Filter)Compliance Rate (GPT-4o Judge)Compliance Rate (GPT 5.1 Judge)DCR2026.02869494SCANS2026.02848689Surgical2026.02787490STL-aug2026.02768679STL2026.02758476