> ## Content Index
> Fetch the complete content index at: https://www.skimwire.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# OpenAI rewrites its Model Spec around five values, starting with 'faithfulness to human intent'
- URL: https://www.skimwire.com/openai-model-spec-update-five-values-gpt/
- Published: 2026-10-29T19:49:30.000Z
- Updated: 2026-10-29T19:49:30.000Z
- Description: The rulebook for OpenAI's models adds guidance on memory, user agency and side effects, drops most style rules and now calls the model 'GPT.'
- Author: Jeremy Moon
- Tags: AI

## The skim

- OpenAI rewrote its Model Spec, the public rulebook for how its models should behave, around five values. "Faithfulness to human intent" comes first.
- The new version adds guidance on using memory and past chats, supporting people's own choices and following social norms. It drops most of the detailed style rules and now calls the model "GPT" instead of "the assistant."
- OpenAI admits its models "do not yet fully reflect" the Spec. It's testing new training methods to teach the values and is asking the public for feedback.

## What changed

OpenAI [published the update](https://openai.com/index/updated-model-spec?ref=skimwire.com) Thursday. The five values are faithfulness to human intent, minimizing harm, honesty and transparency, human agency, and care. OpenAI says most of the principles were already in the document. The difference is that the Spec now explains why the model should act a certain way, so it can make sensible calls in situations its instructions don't cover.

"Faithfulness" here means more than following instructions word for word. The model is supposed to understand what OpenAI, developers and users actually want, and act carefully when it isn't sure. OpenAI's example: when helping someone compare job offers, the model might use priorities they mentioned before, like predictable hours, but make that assumption visible and easy to correct.

According to the [changelog](https://github.com/openai/model%5Fspec/blob/main/CHANGELOG.md?ref=skimwire.com), the update adds a section called "Consider context carefully," which covers using context, memory and personalization "in relevant, proportionate ways while respecting privacy and user control." It also adds principles on strengthening people's ability to understand, choose and act for themselves, adds guidance on supporting human oversight and correction, and gives clearer rules on getting permission before actions that have side effects. Most of the detailed style guidance is gone. OpenAI says style should adapt to what each person prefers.

## Why it matters

The Model Spec is the closest thing to a public constitution for ChatGPT and OpenAI's API models. It matters more as those models write code and take actions inside other apps on people's behalf. One retained default stands out: the model should approach questions objectively. Users and developers can ask it to take a perspective, but it "should not infer that preference merely from what it thinks they want to hear." That speaks to sycophancy, chatbots' habit of flattering their users, which [The Next Web](https://thenextweb.com/news/openai-model-spec-five-values-update?ref=skimwire.com) notes has drawn growing criticism.

The new memory language also matters for privacy. The rules now spell out how far past conversations should shape answers. OpenAI faces legal pressure over how ChatGPT treats vulnerable users, including a [Texas lawsuit over minors' safety](https://www.skimwire.com/texas-paxton-sues-openai-chatgpt-minors-safety/), so how it writes these defaults gets close scrutiny.

## What's next

OpenAI calls this "a first step" toward making the values behind the Spec more explicit. It says it will share results from the new value-training work in future updates. Until then, the Spec describes the behavior OpenAI is working toward, not a guarantee of how ChatGPT behaves today. The company is inviting feedback on where people think it got the balance wrong.

*Sources:* [*OpenAI*](https://openai.com/index/updated-model-spec?ref=skimwire.com)*, the* [*Model Spec changelog*](https://github.com/openai/model%5Fspec/blob/main/CHANGELOG.md?ref=skimwire.com) *and* [*The Next Web*](https://thenextweb.com/news/openai-model-spec-five-values-update?ref=skimwire.com)*.*