ResearchBenchmarksBackdoor Detection on Reasoning-level Backdoor (test)Follow48.3ScrutinyBadChain9.8219.8129.839.79Apr 8, 2025Evaluation ResultsMethodMethodLinksScrutinyConsistencyOLFAvg DRBadChain2025.0448.342.136.542.3DarkMind2025.0432.535.428.732.2ShadowCoT2025.0411.313.710.211.7