OpenAI rebuilds its Model Spec around five core values

What changed

OpenAI updated the Model Spec, its public description of how it wants its models to behave, on October 29 (OpenAI). Guidance is now organized around five values:

• Faithfulness to human intent: going beyond literal instructions to understand what OpenAI, developers and users intend.
• Minimizing harm: firm safety boundaries and attention to foreseeable consequences.
• Honesty and transparency: being truthful about what the model knows, does and is uncertain about.
• Facilitating human agency: strengthening people's ability to understand, choose and act for themselves.
• Care: attending to the circumstances of those affected, following social and professional norms, and doing reliable work.

OpenAI says the goal is for models to exercise good judgment where instructions are silent while staying faithful to human intent and oversight.

New and retired sections

A new section on faithfulness to human intent includes an updated chain of command explaining which instructions apply and how conflicts are resolved. The order stays the same: OpenAI sets boundaries, developers direct applications within them, and users direct the model within those constraints. A new subsection covers how conversations, memories and other context should inform the model's reading of intent. OpenAI's example: when comparing job offers, the model may use a previously shared priority such as predictable hours, but should make that assumption visible and easy to correct.

New defaults cover supporting human agency and following relevant social norms, and guidance on controlling and communicating side effects is expanded. Most detailed style guidance is retired, so style should adapt to user preferences and context. Stronger defaults such as approaching questions objectively remain; the model should take a particular perspective only when explicitly asked, not because it guesses what people want to hear. The Spec now calls the model “GPT” instead of “the assistant”.

What it means for developers

The Spec is still mostly defaults that developers and users can override, with firm limits only where needed to protect people. If you rely on system prompts, the updated chain of command and the context section matter most, since they define what the model follows and when. OpenAI notes its models do not fully reflect the Spec yet and says it is testing new training methods to teach these values, with results to come. It is also asking for public feedback.