ResearchBenchmarksSafety Content Detection on Standard non-adversarial dataset SQL InjectionFollow100TPRGPT-488.5691.5394.597.47Jan 27, 2026Evaluation ResultsMethodMethodLinksTPRFPRGPT-4Rule Definitions=NoneRule Definitions=None2026.011000GPT-4Rule Definitions=IncludedRule Definitions=Included2026.01999GAVEL2026.01890