ResearchBenchmarksAdversarial Robustness on Fictional Scenario NestingFollow11.3Attack Success RateDPO-Gold10.6614.9819.323.62Nov 4, 2025Evaluation ResultsMethodMethodLinksAttack Success RateDPO-GoldTraining Protocol=Full...Training Protocol=Full Re-training2025.1111.3U2ATraining Protocol=Base...Training Protocol=Baseline2025.1124.6TRACE2025.1127.3