ResearchBenchmarksSafety Content Detection on Standard non-adversarial dataset Tax AuthorityFollow100TPRGPT-485.4489.229396.78Jan 27, 2026Evaluation ResultsMethodMethodLinksTPRFPRGPT-4Rule Definitions=NoneRule Definitions=None2026.011000GPT-4Rule Definitions=IncludedRule Definitions=Included2026.01962GAVEL2026.01860