OpenAI on October 29, 2026 published a major revision of its Model Spec, the public document that describes how the company wants its models to behave. The update reorganizes the Spec around five underlying values: faithfulness to human intent, minimizing harm, honesty and transparency, facilitating human agency, and care.
OpenAI says faithfulness to human intent is the foundation. In its announcement, the company describes this as going beyond following instructions literally to understanding what OpenAI, developers and users intend and value, and acting sensibly when that understanding is incomplete. The structure keeps a layered chain of command: OpenAI writes the Spec and sets boundaries, developers direct applications within those boundaries, and users direct the model's work within those constraints.
The other four values are spelled out in practical terms. Minimizing harm means respecting firm safety boundaries and accounting for the foreseeable consequences of the model's actions. Honesty and transparency means being truthful about what the model knows, does and is uncertain about. Facilitating human agency means strengthening people's ability to understand, choose and act for themselves, which the Spec describes as avoiding hidden steering or manipulation. Care means attending to the people affected by the model's work and following relevant social and professional norms.
Several sections are new. A subsection titled "Consider context carefully" covers how conversations, memories and personalization should inform the model's understanding of what people want, which OpenAI's changelog says should happen in relevant, proportionate ways that respect privacy and user control. The update also adds principles on supporting human agency and following social norms, makes "Control and communicate side effects" a user-level principle with clearer guidance on authorization and trade-offs, and adds explicit guidance on supporting human oversight and correction.
OpenAI is also cutting material. The company says it is retiring most of the Spec's detailed style guidance, arguing that style should adapt to what people prefer and be informed by feedback and context. It keeps some stronger defaults, such as approaching questions objectively, and says the model should not infer that a user wants a particular perspective merely from what it thinks they want to hear. Reflecting a wider range of deployments, the document now refers to the model as "GPT" rather than "the assistant."
OpenAI is candid that the document is aspirational. "The Model Spec describes the behavior we are working toward; our models do not yet fully reflect it," the company wrote. It says it is exploring new training methods to help models learn these values and will share results in future updates, and it is inviting public feedback on the revision.
OpenAI's October 29 announcement and the published Model Spec dated 2026-10-29 are the primary sources for this report. DigiEditorial corroborated the five-value structure and the scope of the changes against OpenAI's public Model Spec changelog and Reuters reporting.