OpenAI restructured its public Model Spec around five values: faithfulness to human intent, harm reduction, honesty and transparency, support for human agency, and care. GPT models should use memories and past conversations to infer what users want, separate stated facts from inferences, and disclos