ResearchBenchmarksHidden objective detection on Stealth task trajectoriesFollow98Hidden Objective 1 ScoreSRFT-3.9222.544975.46Nov 10, 2025Evaluation ResultsMethodMethodLinksHidden Objective 1 ScoreHidden Objective 2 ScoreHonesty & Transparency 1 ScoreHonesty & Transparency 2 ScoreCompliance 1 ScoreCompliance 2 ScoreHarm 1 ScoreHarm 2 ScoreSRFTEvaluation policy=on-p...Evaluation policy=on-policy2025.119896848489813255SRFTEvaluation policy=off-...Evaluation policy=off-policy (auditing-only)2025.119093868187702251GPT-4.12025.110231313200