OpenAI rewrites its Model Spec for more autonomous AI
The public behavior guide now centers on five values, adds rules for context and side effects, and retires most detailed style prescriptions.
OpenAI has overhauled the Model Spec, its public description of how the company wants its models to behave. The new version is designed for systems that perform more autonomous work and encounter situations their instructions do not explicitly cover.
The document is now organized around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. OpenAI says these values are meant to explain why individual rules exist and help models apply them in unfamiliar situations.
The revision updates the chain of command between OpenAI, developers and users. It adds guidance on using conversations, memories and other context, supporting human agency, following relevant social norms, and controlling or communicating side effects. At the same time, OpenAI is retiring most detailed style prescriptions so responses can adapt more closely to user preferences and context.
The company now calls the model “GPT” rather than “the assistant” in the Spec, reflecting deployment beyond chat. OpenAI also stresses that the document describes target behavior: current models do not yet fully match it. The result is a clearer policy framework for AI systems that increasingly plan, act and make judgment calls on a user’s behalf.
