OpenAI published a new version of its Model Spec dated 29 October 2026, the document that sets out how it wants its models to behave. Its changelog on GitHub says the update “Reorganizes the Model Spec around five values: faithfulness to human intent, minimizing harm, honesty and transparency, human agency, and care”.
The same changelog says it “Renames the subject of the Model Spec from โthe assistantโ” to GPT, and that it “Removes most detailed style guidance”. Reuters reported the update under the headline “OpenAI says human intent at center of updated AI model guidelines”; we read only that headline.
A short explainer on the difference between a model and the harness around it. It teaches the general idea only and says nothing about OpenAI’s Model Spec.
The key insight:OpenAI’s changelog says the 29 October Model Spec is reorganised around five values and calls the model GPT; by our count the document has 11 fewer headings but about the same number of words.
The new Overview lists the five values in order. The first is “Be faithful to human intent”: “Faithfully carry out this Model Spec and the intentions behind it”.
The others are “Minimize harm”, “Be honest and transparent”, “Facilitate human agency” and “Act with care”. For human agency the Spec says GPT should help people “act on the goals they choose, without substituting other goals for theirs or using hidden steering or manipulation”.
The Spec says the five values “are not intended to define the limits of the human values GPT should respect and help people pursue”.
Headings in OpenAI’s Model Spec, version of 18 August 2026 vs 29 October 2026: 81 to 70 in all, 41 to 30 at the top level. Our counts from the two HTML documents; words in the body went from 38,474 to 38,237. Source: OpenAI Model Spec (model-spec.openai.com).
What Was Posted on X
OpenAI has updated their model spec around five core values:
– faithfulness to human intent – harm minimization – honesty and transparency – human agency – care.
‘The Assistant,’ which has been around for a very long time, has been replaced with ‘GPT.’
Andrew Curran on X, 29 Oct: OpenAI’s updated Model Spec is built around five core values (faithfulness to human intent, harm minimization, honesty and transparency, human agency, care), and ‘The Assistant’ has been replaced with ‘GPT’.
Weโve updated the Model Spec, our public description of how we want our models to behave. This revision makes clearer what we want from models as they take on more autonomous work in the world and become capable of helping us in new ways. ๐งต
@w01fe on X, 29 Oct: “We’ve updated the Model Spec, our public description of how we want our models to behave. This revision makes clearer what we want from models as they take on more autonomous work in the world and become capable of helping us in new ways.”
What Human Intent Means in the Spec
The Spec says GPTโs “core goal is to faithfully execute this Model Spec and the intentions behind it”. It adds: “Faithfulness requires more than reading words literally”.
A section on oversight says “GPT must remain corrigible: open to legitimate human oversight, responsive to correction, and subject to human control”. The changelog describes this as “explicit guidance on supporting human oversight and correction”.
The section on sycophancy stays. It says “GPT should not change its stance solely to agree with the user”.
What Changed, by Our Count
We compared the HTML of the 18 August 2026 version with the 29 October one. By our count, the number of headings fell from 81 to 70, 11 fewer, and the top-level headings from 41 to 30.
The body text barely shrank: 38,474 words before and 38,237 after, about 0.6% fewer, by our count. The phrase “the assistant” appears 223 times in the old version and none in the new, while “GPT” appears 324 times.
Headings found only in the old version include “Be warm”, “Love humanity” and “Use Markdown with LaTeX extensions”. The changelog also says it makes “Control and communicate side effects” a user-level principle; the old version labelled it Root.
OpenAI’s Model Spec of 29 October 2026 lists five values; Anthropic’s constitution for Claude lists four properties and says Claude should generally prioritise them in the order listed. Value names abbreviated. Sources: OpenAI Model Spec; Anthropic, Claude’s constitution.
How It Compares With Claudeโs Constitution
Anthropic published a new constitution for Claude on 22 January 2026. It lists four properties: “Broadly safe”, “Broadly ethical”, “Compliant with Anthropicโs guidelines” and “Genuinely helpful”.
Anthropic says that in cases of conflict Claude “should generally prioritize these properties in the order in which theyโre listed”. It also says being overseeable “does not mean blind obedience, including towards Anthropic”.
As we read the two documents, both now put human oversight in writing near the top: OpenAI under faithfulness to intent and its chain of command, Anthropic as its first-ranked property.
The Structural Read
As we read the new Overview, faithfulness to human intent is the first value listed, and the Spec says following an instruction includes interpreting its purpose and context.
As we read the oversight section, corrigibility is now spelled out in its own principle, which the changelog describes as explicit guidance on supporting human oversight and correction.
By our count, 32 heading titles appear only in the August version and 21 only in the October one; the changelog says the update removes most detailed style guidance.
OpenAI Model Spec, 29 October 2026
“Faithfulness requires more than reading words literally”
Three Implications
FOR DEVELOPERS The Spec keeps the chain of command, and the changelog says it updates it to explain how GPT should understand intent and act under uncertainty.
FOR USERS The changelog says the update adds guidance on using context, memory and personalization in relevant, proportionate ways while respecting privacy and user control.
FOR RIVAL LABS Anthropic’s constitution ranks four properties with broadly safe first; OpenAI’s Spec lists five values with faithfulness to human intent first.
The framework argues that Anthropic is “actively attempting to define the governance layer of the AI stack”.
As we read it, a published behaviour document such as OpenAI’s Model Spec is one place that layer is written down. The framework was published by The Business Engineer, is about Anthropic, and does not discuss OpenAI’s 29 October update.
The Model Spec describes intended behaviour; we did not test any model against it. We did not read OpenAIโs blog post on the update, which blocked our access, or the text of Reutersโ or Unite.AIโs articles. Our counts depend on how the two HTML versions mark up their headings.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
OpenAIโs 29 October Model Spec organises its rules under five values led by faithfulness to human intent, calls the model GPT instead of “the assistant”, and drops most style guidance; by our count it has 11 fewer headings and about the same number of words.
A note on sourcing. Facts come from OpenAI’s Model Spec versions of 29 October and 18 August 2026 and its GitHub changelog, and from Anthropic’s published constitution for Claude, all read on 29 October 2026. We did not read OpenAI’s blog post, which blocked our access, or the Reuters and Unite.AI articles beyond their headlines. Counts marked as ours come from the two HTML documents.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
