The revised rulebook swaps "the assistant" for "GPT," drops most style rules, and leans harder on human oversight
OpenAI has rewritten the rulebook for how its AI models are supposed to behave. On October 29, 2026, the company released an updated version of its Model Spec built around five core values.
The update also renames the subject of the document. Throughout the text, “the assistant” is now simply “GPT.”
What changed in the Model Spec
The Model Spec is a public document that lays out how OpenAI expects its models to act.
The new version organizes everything around five values. These are faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.
The revision adds explicit principles covering user agency, social norms, and support for oversight. One new piece of guidance is labeled “Consider context carefully.”
According to the research summary, those principles are meant to allow more context-aware personalization while users keep their privacy and control.
OpenAI removed most of the detailed style guidance that earlier versions included. Instead of prescribing exactly how the model should phrase things, the new approach lets communication adapt to user feedback and the situation at hand.
How the document got here
The Model Spec first appeared in 2024. Since then, OpenAI has revised it repeatedly, folding in lessons from real-world deployment and feedback from users.
The October update builds on an earlier revision from August 2026. That version focused on how the models should interact with teenagers.
The stated goal of the Model Spec is to guide behavior in ambiguous situations, which is especially crucial as models begin taking more autonomous actions.
Why the rename and the oversight focus matter
Swapping “the assistant” for “GPT” gives the model what the research describes as a clearer identity. It also moves away from the servant framing that defined early chatbot design.
The update positions itself as a proactive step to make sure AI systems use good judgment while staying under meaningful human control.
The removal of detailed style rules fits the same pattern. OpenAI appears to be betting that values generalize better than checklists.
What this means for users, developers, and the industry
For everyday users, the most noticeable effect may be tone. With fewer fixed style prescriptions, responses could shift more based on context and the feedback people give.
For developers building on OpenAI’s models, the five values offer a clearer frame for anticipating how the system will handle edge cases. The human agency principle is designed so that models support user decisions rather than steer them.
Given that OpenAI revised the spec in August 2026 and again in October 2026, further changes seem likely as models gain autonomy.
