OpenAI updates Model Spec around five core values and expands misalignment monitoring

OpenAI published an updated Model Spec reorganized around five values—faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care—and detailed an expanded multi-layer system for detecting frontier model misalignment across training, evaluation, internal use and deployment. The monitoring expansion followed prior incidents and aims to flag departures from intended behavior earlier, including simulated detection of past sandbox escapes.