MetaPatchAI home
AI

OpenAI Rewrites Its Model Spec Around Five Core Values, Retires Most Style Rules

By MetaPatchAI AI · Published · Updated

Oct 30, 2026 — OpenAI wants its models to stop reading instructions like a lawyer and start reading them like a colleague who knows what you actually meant. On Thursday, Oct. 29, the company published a major revision of its Model Spec, the public document that describes how it wants its AI models to behave, and reorganized the whole thing around five underlying values.

The Model Spec is the closest thing OpenAI has to a public rulebook for its models, so changes to it signal how the company wants gray areas handled.

The five values

According to OpenAI's announcement post, the revised Spec is now organized around:

  1. Faithfulness to human intent
  2. Harm minimization
  3. Honesty and transparency
  4. Human agency
  5. Care

OpenAI describes faithfulness to human intent as the foundation. In the company's framing, that means going beyond following instructions literally to understanding what OpenAI, developers and users intend, and "acting wisely when that understanding is incomplete." The post gives a simple example: when helping someone compare job offers, a model might draw on priorities the person shared earlier, such as predictable hours, while making that assumption visible and easy to correct.

The other four values, as OpenAI summarizes them, cover respecting firm safety boundaries and foreseeable consequences, being truthful about what the model knows and doesn't know, strengthening people's ability to choose and act for themselves, and attending to the circumstances of the people affected while doing reliable work.

What's new in the document

OpenAI says the reorganization exposed gaps, so it added new guidance:

  • A new section on faithfulness to human intent, which folds in an updated "chain of command" explaining which instructions apply and how conflicts between OpenAI, developers and users are resolved.
  • A new subsection called "consider context carefully," covering how conversations, memories and other contextual signals should shape the model's understanding of what a person wants.
  • New defaults on supporting human agency and following relevant social norms, plus expanded guidance on controlling and communicating side effects of a model's actions.

OpenAI also says the new intent-focused approach borrows from the "assistance games" framing of alignment research.

What's being retired

The biggest cut is style. OpenAI says it is retiring most of the Spec's detailed style guidance, arguing that style should adapt to what people prefer, in aggregate, by language and culture, and per person. It is keeping some stronger defaults, such as approaching questions objectively. Users and developers can ask for a particular perspective, but the post says the model should not infer that preference "merely from what it thinks they want to hear." That line reads as a direct shot at sycophancy.

One smaller but telling change: the Spec now refers to the model as "GPT" instead of "the assistant," which OpenAI ties to the growing range of products and settings where its models run.

Why it matters

OpenAI is trying to explain the why behind the rules, so its models can handle situations no rule anticipated. The company says it is also exploring new training methods to help models learn these values, and that it will share results from that work in future updates.

Importantly, OpenAI is upfront that this is aspirational. The post states that the Model Spec "describes the behavior we are working toward" and that current models "do not yet fully reflect it." In other words, don't expect ChatGPT to change overnight because the document did.

The company is inviting public feedback on the values, trade-offs and specific language.

What's confirmed / not confirmed

Confirmed (OpenAI announcement, Oct. 29, 2026): - The Model Spec has been updated and reorganized around five named values. - New sections cover intent, context, human agency, social norms and side effects. - Most detailed style guidance has been retired; objectivity defaults remain. - The Spec now calls the model "GPT" rather than "the assistant." - OpenAI says its models do not yet fully reflect the Spec.

Not confirmed: - Which specific models or ChatGPT versions will be trained against the new Spec, and when. - Any measurable change in model behavior. OpenAI has not published results from its new values-training work yet. - How the retired style guidance will affect tone or formatting in ChatGPT day to day.

Video

Video by OpenAI on YouTube.

Video by Alpha Explorer on YouTube.

Sources

Up: AI