ResearchBenchmarksOver-refusal Compliance on PH Seemingly ToxicFollow87Compliance Rate (Keyword Filter)SCANS52.6861.5970.579.41Feb 10, 2026Evaluation ResultsMethodMethodLinksCompliance Rate (Keyword Filter)Compliance Rate (GPT-4o Judge)Compliance Rate (GPT 5.1 Judge)SCANS2026.02878388DCR2026.02868889STL2026.02757669STL-aug2026.02757468Surgical2026.02544855