OpenAI released a new version of its "Model Spec" today, restructuring the guidance framework for GPT behavior around five core values: faithfulness to human intent, harm reduction, honesty and transparency, preservation of human autonomy, and care for affected people. The updated spec introduces guidance on contextual understanding, requiring models to interpret users' true goals and long-term interests rather than executing instructions literally, while explicitly prohibiting steering user choices without consent. The revision also adjusts instruction priority rules and default behaviors, removing most detailed writing style specifications to allow models to adapt expression based on user preferences, language, culture, and context. OpenAI acknowledged that currently deployed models do not yet fully meet the new behavioral standards, stating the company will continue improving them through training and real-world deployment feedback.