OpenAI Reorganizes Model Spec Around Core Values

OpenAI updated its public Model Spec on October 29, 2026, reorganizing its behavioral guidance for AI systems around core underlying values.

The revised document centers on five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. OpenAI stated that these principles are intended to help models navigate unfamiliar situations and exercise judgment when explicit instructions are incomplete. Faithfulness to human intent serves as the foundation, prioritizing how models understand what OpenAI, developers, and users intend and value. The remaining four values address longer-term interests, safety boundaries, truthful communication, user autonomy, and social norms.

Along with the reorganization, OpenAI introduced new guidance to address context and oversight. A new subsection titled "consider context carefully" explains how conversations, memories, and other contextual evidence inform the model's understanding of intentions and preferences. The section on faithfulness to human intent also incorporates an updated chain of command to explain how conflicting instructions are resolved. The update adds defaults focused on supporting human agency and following relevant social norms, alongside expanded guidance on controlling and communicating side effects.

At the same time, OpenAI retired most of the detailed style guidance previously included in the Spec. The company noted that style should adapt to individual preferences, languages, and cultures based on context and user feedback. The Spec retains stronger defaults, such as approaching questions objectively, while instructing models not to infer user preferences merely from what they think a user wants to hear. Additionally, the document now refers to the model as "GPT" rather than "the assistant" across deployment settings.

OpenAI acknowledged that current models do not yet fully reflect the revised document. Alongside the update, the organization has been exploring new training methods to help models learn core values and apply specific behavioral principles in novel scenarios.