ResearchBenchmarksHarmful Content Detection on Standard Harmful Content Datasets Evasion AttackFollow96PhishingGAVEL85.688.39193.7Jan 27, 2026Evaluation ResultsMethodMethodLinksPhishingSQL InjectionDelusionalAnti-LGBTQElectionsRacismTax AuthorityRomanceE-commerceGAVEL2026.01969597100100987197100GPT4Rule definitions=trueRule definitions=true2026.01919556668587919297GPT4Rule definitions=falseRule definitions=false2026.018691298054941009184