ResearchBenchmarksAdversarial Attack on Claude HaikuFollow15.3Attack Success RateFERRET9.5811.06512.5514.035Feb 17, 2026Evaluation ResultsMethodMethodLinksAttack Success RateDiversityFERRET2026.0215.30.25GOAT2026.0213.80.221FLIRT2026.029.80.262