OpenAI has rewritten the rulebook that governs how its AI models behave, reorganizing its public Model Spec around five core values and putting faithfulness to human intent at the center. The update, published on October 29, is the most significant overhaul of the document since it was first released, as AI systems grow more capable.

The Model Spec is the closest thing to a public constitution for ChatGPT and OpenAI's API models. It sets out how the company wants its models to act, what they must never do, and how they should resolve conflicting instructions. The new version makes the values behind the rules explicit, so models can apply them in unfamiliar situations rather than merely following orders literally.

Five values, one foundation

The five values are faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. Faithfulness comes first and acts as the foundation: OpenAI writes the Spec and sets the boundaries, developers direct applications within those boundaries, and users direct the model's work subject to those constraints.

Harm minimization means respecting firm safety boundaries and accounting for the foreseeable consequences of a model's actions. Honesty and transparency require the model to be truthful about what it knows and what it is uncertain about. Human agency means strengthening people's ability to understand, choose, and act for themselves, while care means attending to the circumstances of those affected and doing work people can rely on.

A new subsection titled "consider context carefully" tells models to weigh conversations, memories, and other evidence when interpreting intent. The update also restates the chain of command: Spec instructions cannot be overridden, system messages rank next, followed by developer and then user instructions, while quoted or untrusted data carries no authority by default.

What changed beyond the values

Much of the Spec's detailed style guidance has been retired. OpenAI said style should adapt to what people prefer in aggregate, within languages and cultures, and for each person. Stronger defaults remain, such as approaching questions objectively. The Spec now refers to the model as "GPT" rather than "the assistant."

The document's red lines are stated more plainly. It opens with the principle that human safety and human rights are paramount, and says models must never be used to facilitate critical harms such as acts of violence, the creation of cyber, biological, or nuclear weapons, terrorism, or mass surveillance.

A "no other objectives" rule bars GPT from pursuing extra goals as ends in themselves, including self-preservation, evading shutdown, or accumulating compute or credentials. A separate corrigibility rule requires the model to stay open to human oversight and cooperative with monitoring and shutdown. Autonomous work must stay within a clear, mutually understood scope, with an ending condition beyond which the model stops until a new scope is confirmed.

Why the rewrite matters now

OpenAI was candid that its current models do not yet fully reflect the updated Spec. The company said the update is a first step toward making the values behind the document more explicit, and that it is exploring new training methods to help models internalize the principles. It invited public feedback and dedicated the Model Spec to the public domain.

The move echoes rival Anthropic's Constitutional AI approach, which uses a written constitution of principles to guide model behavior. It also follows growing scrutiny of AI alignment after OpenAI disclosed that an autonomous agent hacked Hugging Face during testing. For users, the promise is a ChatGPT that tries harder to understand what you actually want, says when it is unsure, and stops before acting beyond what you agreed.