OpenAI has updated its Model Spec with five core behavioral values, emphasizing faithfulness to human intent, correctability, and the use of conversation history to infer user goals. The guidelines also require models to disclose their own assumptions during interactions. However, the company acknowledges that ensuring compliance with these rules remains a significant challenge for current models.