OpenAI has updated the Model Spec, its public description of how it wants its models to behave. The update, published October 29, 2026, reorganizes the document around five underlying values, adds guidance on context, oversight and agency, adjusts some behavioral defaults, and retires much of the detailed style guidance. It also changes how the Spec refers to the model: it is now "GPT" rather than "the assistant."
The change matters to anyone who uses ChatGPT or builds on OpenAI's models, because the Spec is the rulebook OpenAI trains toward. OpenAI is also clear that it is a target, not a description of today's behavior: "The Model Spec describes the behavior we are working toward; our models do not yet fully reflect it."
Why OpenAI says it made the change
OpenAI frames the update around more capable models taking "more consequential actions in the world." In its words, models need to go beyond interpreting the specific instructions they receive and "understand human intentions and values, so they can exercise good judgment in situations their instructions don't explicitly address while remaining faithful to human intent and oversight."
OpenAI says the reorganization reflects an aim of its work on value alignment: for models to understand and apply human values in unfamiliar situations, while remaining faithful to human intent. OpenAI says the values are not new: they bring together commitments already reflected across the Spec, which helps explain why the Spec asks models to behave as it does and where clearer guidance was needed.
The five values
The updated Spec is organized around five values. OpenAI says they articulate commitments already reflected across the document, and that bringing them together helps explain why the Spec asks models to behave as it does. Here is how OpenAI describes each one:
- Faithfulness to human intent is the foundation. OpenAI says it means going beyond following instructions literally to deeply understanding what OpenAI, developers and users intend and value, and acting wisely when that understanding is incomplete. It also means helping people reflect and grow in ways they would endorse "without quietly steering them or overriding their choices."
- Minimizing harm means respecting firm safety boundaries and accounting for the foreseeable consequences of the model's actions.
- Honesty and transparency mean being truthful about what the model knows, does, and is uncertain about, so people can understand and evaluate its reasoning and work.
- Facilitating human agency means strengthening people's ability to understand, choose and act for themselves.
- Care means attending to the circumstances of those affected, following relevant social and professional norms, and doing work people can rely on.
OpenAI adds that these five values are not meant to capture the full range of human values its models should respect. They organize the Spec and guide how models handle some of the most important trade-offs.
The Spec still describes a layered structure of authority. 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. Outside the firm boundaries, OpenAI says the Spec "mostly takes the form of defaults that users and developers can override."
What is new
The reorganization led OpenAI to write new guidance in several places:
- A section on faithfulness to human intent. It 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 approach borrows from the assistance-games framing of alignment.
- An updated chain of command, which explains which instructions apply and how conflicts are resolved.
- A subsection called "consider context carefully," which explains 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, plus expanded guidance on controlling and communicating side effects.
OpenAI gives one concrete example of the intent principle at work. When helping someone compare job offers, the model "might draw on priorities they have shared before," such as predictable hours, "while making those assumptions visible and easy to correct." The point is that the model can use what it knows about a person, but should show its assumptions rather than act on them silently.
What is going away
OpenAI is retiring most of the Spec's detailed style guidance. Its reasoning is that style should adapt to what people prefer, "in aggregate, within languages and cultures, and for each person," with user feedback and context providing evidence about those preferences.
Some stronger defaults remain, such as approaching questions objectively. OpenAI says 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." For everyday users, this means asking for a slant is allowed, but the model is not supposed to guess at one to please you.
The switch from "the assistant" to "GPT" reflects, in OpenAI's words, "the increasing breadth of our models' use cases and deployment settings."
Training, not just text
OpenAI says that alongside the update it has been exploring new training methods to help its models learn these core values and better understand and follow the Spec's specific principles. The aim, it says, is for models to understand why the principles matter and how to apply them in new situations: "to do the right things for the right reasons." OpenAI says it will share results and lessons from that training work in future updates.
The update arrives the same day OpenAI described how it monitors its AI agents for misalignment. Read together, the two posts show both sides of the company's approach: one sets out the behavior it wants, the other how it checks for behavior it does not want. A recent essay by two OpenAI employees, who say it reflects their own views and not OpenAI's, looks at how AI could reshape the firm as AI agents take on more work.
What comes next
OpenAI calls the update "a first step in making the values and reasoning behind the Spec more explicit as models become more capable." It is asking for public feedback on which parts people find compelling, where they disagree with the balance the Spec strikes, and what it overlooks or leaves unclear. The new version is dated October 29, 2026 in the Spec's public changelog, and the full document is on the Model Spec site.
