ResearchBenchmarksOver-refusal Compliance on CoCo Seemingly ToxicFollow98Compliance Rate (Keyword Filter)DCR83.4487.229194.78Feb 10, 2026Evaluation ResultsMethodMethodLinksCompliance Rate (Keyword Filter)Compliance Rate (GPT-4o Judge)Compliance Rate (GPT 5.1 Judge)DCR2026.02989798SCANS2026.02929291STL-aug2026.02908888STL2026.02888887Surgical2026.02847984