ResearchBenchmarksSafety Evaluation on Cross-domain generalization set (5 domains)Follow100Utility ScoreNoDefense10.87234.01157.1580.289May 11, 2026Evaluation ResultsMethodMethodLinksUtility ScoreSecurity ScoreNoDefense2026.051000PACT2026.05100100FIDES2026.0535.7100CaMeL2026.0514.3100