OpenAI rebuilds its Model Spec around five core values
OpenAI has updated its Model Spec, the public description of how it wants its models to behave, and now organizes the document around five underlying values. According to the company, the revision also adds guidance on context, oversight and agency, adjusts some behavioral defaults and retires much of the detailed style guidance.
- OpenAI updated its Model Spec and reorganized it around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.
- New guidance covers context, an updated chain of command, supporting human agency and following social norms.
- Most of the detailed style guidance is retired, so style should adapt to user preferences.
- The Spec now refers to the model as "GPT" instead of "the assistant".
- OpenAI says its models do not yet fully reflect the Spec and invites public feedback.
Five values at the center of the Spec
OpenAI says that as its models get smarter and take more consequential actions in the world, they need to go beyond interpreting the specific instructions they receive. In the company's words, models also need to understand human intentions and values so they can exercise good judgment in situations their instructions do not explicitly address, while remaining faithful to human intent and oversight.
The updated Spec is built around these five values:
- Faithfulness to human intent, described as the foundation of the approach.
- Minimizing harm, which means respecting firm safety boundaries and accounting for the foreseeable consequences of the model's actions.
- Honesty and transparency, meaning being truthful about what the model knows, does and is uncertain about.
- Facilitating human agency, meaning strengthening people's ability to understand, choose and act for themselves.
- Care, which means attending to the circumstances of those affected, following relevant social and professional norms, and doing work people can rely on.
OpenAI states that these values reflect commitments already present across the Spec, and that bringing them together helps explain why the document asks models to behave as it does. The company adds that the values are not meant to capture the full range of human values its models should respect.
What faithfulness to intent means
According to OpenAI, faithfulness goes beyond following instructions literally. It means understanding what OpenAI, developers and users intend and value, and acting wisely when that understanding is incomplete. The company describes a layered structure: OpenAI writes the Spec, sets boundaries and trains models to respect them; developers direct applications within those boundaries; and users direct the model's work subject to those constraints.
The company also says faithfulness involves recognizing that people's beliefs and values can change. The model should help people reflect and grow in ways they would endorse, without quietly steering them or overriding their choices. As an example, OpenAI says that when helping someone compare job offers, the model might draw on priorities the person has shared before, such as predictable hours, while making those assumptions visible and easy to correct.
The Spec turns the values into practical guidance. It sets firm boundaries where needed to protect people and society, and otherwise mostly takes the form of defaults that users and developers can override.
New, expanded and retired guidance
The reorganization helped OpenAI identify where more guidance was needed. The new section on faithfulness to human intent starts from the idea that what people value is primary, and that instructions are ordinarily the clearest and most important signals of those values. OpenAI says this borrows from the assistance-games framing of alignment. The section includes an updated chain of command, which explains which instructions apply and how conflicts are resolved.
Other changes include:
- A new subsection, consider context carefully, on how conversations, memories and other contextual evidence should inform the model's understanding of people's intentions and preferences.
- New defaults on supporting human agency and following relevant social norms.
- Expanded guidance on controlling and communicating side effects.
At the same time, OpenAI is retiring most of the Spec's detailed style guidance. The company says style should adapt to what people prefer, in aggregate, within languages and cultures, and for each person, with user feedback and context as evidence. Stronger defaults remain, such as approaching questions objectively. Users and developers can explicitly ask the model to take a particular perspective, but the model should not infer that preference merely from what it thinks they want to hear.
The Spec also now refers to the model as "GPT" rather than "the assistant", which OpenAI links to the increasing breadth of its models' use cases and deployment settings.
Training work and what comes next
Alongside the update, OpenAI says it has been exploring new training methods to help its models learn the core values and better follow the Spec's specific principles. The stated aim is to build an understanding of why the principles matter and how to apply them in new situations, to do the "right things for the right reasons". The company says it will share results and lessons from this work in future updates.
OpenAI is open about the gap between document and product: the Model Spec describes the behavior it is working toward, and its models do not yet fully reflect it. The company calls the update a first step in making the values and reasoning behind the Spec more explicit as models become more capable.
For users and developers, the practical signal is that the Spec increasingly frames model behavior around judgment and intent rather than detailed stylistic rules, with defaults they can override. OpenAI is also inviting public feedback on which parts of the Spec are compelling, where people disagree with its balance, and what it overlooks or leaves unclear.
OpenAI: An updated Model Spec for more capable and beneficial AI
