OpenAI Model Spec Update Rebuilds AI Rules Around Five Values
OpenAI on Thursday, October 29, 2026, published a major update to its OpenAI Model Spec, the public document that describes how its AI models should behave. The revision reorganizes the rules around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.
The company says the change is driven by models that are getting smarter and taking “more consequential actions in the world.” Instead of only following instructions literally, OpenAI wants its models to understand what people actually intend and to use good judgment in situations their instructions do not cover.
What Changed in the OpenAI Model Spec
The Model Spec has existed since 2024 as OpenAI’s public rulebook for ChatGPT and its API models. According to OpenAI’s announcement, the new version groups its guidance under these five values:
- Faithfulness to human intent: understanding what OpenAI, developers and users want, not just the literal wording of a request.
- Minimizing harm: respecting firm safety boundaries and weighing the foreseeable consequences of the model’s actions.
- Honesty and transparency: being truthful about what the model knows, does and is unsure about.
- Human agency: strengthening people’s ability to understand, choose and act for themselves.
- Care: paying attention to the circumstances of those affected and producing work people can rely on.
Faithfulness to human intent is described as the foundation. OpenAI sets the boundaries, developers direct applications within them, and users direct the model’s work within those limits. This “chain of command” section has been updated to explain which instructions apply and how conflicts are resolved.
New Rules on Context, Memory and Side Effects
A new subsection tells models to “consider context carefully,” meaning conversations, memories and other signals should shape how a model reads a person’s preferences. OpenAI gives the example of comparing job offers: the model might recall that a user values predictable hours, but it should make that assumption visible and easy to correct.
The update also adds defaults on supporting human agency and following social norms, and it expands guidance on controlling and communicating side effects. That last part matters more as AI agents book, buy and edit things on a user’s behalf, a trend we have seen in tools like ChatGPT’s iCloud calendar and email integration.
Style Rules Retired, “the Assistant” Becomes “GPT”
OpenAI is retiring most of the Spec’s detailed style guidance. The company says style should adapt to what people prefer, by language, culture and individual user. One stronger default stays: models should approach questions objectively and should not tell people what they think they want to hear unless explicitly asked to take a perspective.
The document also stops calling the model “the assistant” and now refers to it as “GPT,” reflecting how widely the models are now deployed beyond chat.
OpenAI researcher Jason Wolfe, who works on the Spec, announced the update on X:
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. 🧵
— Jason Wolfe (@w01fe) October 29, 2026
What It Means for ChatGPT Users
OpenAI is clear that the Spec describes the behavior it is working toward and that “our models do not yet fully reflect it.” The company says it is testing new training methods to teach models these values and the reasoning behind them, with results to be shared later.
For everyday users, the practical effects should show up gradually: answers that adapt more to personal style, more explicit handling of remembered preferences, and agents that explain side effects before acting. The update comes as AI safety is under heavier scrutiny, including a lawsuit against OpenAI over ChatGPT advice, and as rivals like Mistral push powerful open-weight models such as Mistral Large 4.
OpenAI is inviting public feedback on the values, trade-offs and wording, and you can read the full document on the official Model Spec site.
Your comment has been submitted,
it will be published after approval.