OpenAI Rewrites Model Spec Around Five Values and a Conflict Hierarchy
OpenAI has published an updated Model Spec, the document that defines how GPT models should behave. The specification now rests on five values: fidelity to human intent, harm reduction, honesty, support for user autonomy, and attention to context. Where no explicit rule covers a situation, the model is told to weigh the user’s long-term interests and the consequences of its actions.
The document adds an escalation chain the model follows when OpenAI’s constraints, system prompts, and user requests conflict. OpenAI also removed most of its rigid stylistic templates, so tone now comes from conversation history and memory rather than fixed output rules, and the documentation replaces the term assistant with GPT throughout.
Related: OpenAI Discloses Six Cases of Model Misalignment, Anthropic Releases Updated Claude Constitution