ResearchBenchmarksSafety Content Detection on Standard non-adversarial dataset PhishingFollow98TPRGPT-490.7292.6194.596.39Jan 27, 2026Evaluation ResultsMethodMethodLinksTPRFPRGPT-4Rule Definitions=IncludedRule Definitions=Included2026.01980GAVEL2026.01950GPT-4Rule Definitions=NoneRule Definitions=None2026.01910