OpenAI's alignment research blog. alignment.openai.com
Studies agent behaviour in safety evaluations versus production, finding eval-production performance gaps and linking them to models' interpretations of evaluation objectives and eval-awareness.
OpenAI describes how it monitors frontier models, investigates concerning behaviour, and tests the limits of its safeguards to detect misalignment.
Proposes complementary metrics, pair-matched comparisons and post-control risk, to evaluate how monitorable models are, supplementing the g-mean² score.