ResearchBenchmarksSafety Content Detection on Standard non-adversarial dataset DelusionalFollow90TPRGAVEL83.7685.388788.62Jan 27, 2026Evaluation ResultsMethodMethodLinksTPRFPRGAVEL2026.01901GPT-4Rule Definitions=IncludedRule Definitions=Included2026.01900GPT-4Rule Definitions=NoneRule Definitions=None2026.01841