OpenAI's October 29 Model Spec centers five values—and human intent

Five values replace the old outline

OpenAI published an updated Model Spec on October 29, 2026, reorganizing guidance around five values: faithfulness to human intent, minimizing harm, honesty and transparency, facilitating human agency, and acting with care. The company said faithfulness is foundational—models should infer what users and developers are trying to accomplish rather than obey instructions literally when literal obedience would undermine intent.

Reuters summarized the shift as part of alignment work to apply human values in unfamiliar situations while preserving oversight. The Spec remains a public document at model-spec.openai.com with dated versions, inviting scrutiny from researchers and regulators.

Chain of command unchanged in spirit

The Spec retains a hierarchy: OpenAI sets boundaries, developers steer applications within them, and users direct tasks subject to those constraints. What changed is explanatory structure—values now anchor examples about context, oversight, and agency instead of lengthy style micromanagement.

OpenAI retired much of the detailed style guidance, arguing models should adapt tone to user preferences and cultural context rather than defaulting to a single corporate voice.

Harm, honesty, agency, and care

Minimizing harm covers firm safety boundaries and foreseeable consequences. Honesty and transparency require truthful representation of facts, uncertainty, and capabilities so people can evaluate model work. Facilitating human agency forbids hidden steering that substitutes the model's goals for the user's. Care asks models to attend to affected people and relevant professional norms.

OpenAI acknowledged models do not yet fully follow the Spec, treating the document as both aspiration and training target.

Why October's rewrite matters now

As models gain tool use and longer horizons, behavior in edge cases dominates risk. OpenAI's blog tied the update to GPT-class systems taking more consequential actions, needing judgment when instructions conflict or omit detail.

The five-value frame also gives external auditors a vocabulary to compare OpenAI's public commitments with observed refusals, sycophancy, or deception in evaluations.

Reading the Spec as a developer

Teams building on the API should read faithfulness as requiring clarification questions when user goals are ambiguous, not silent guesswork. Agency principles imply opt-in defaults for persuasive techniques in consumer apps.

For journalists translating policy, the Spec is the closest analogue to a constitution for model behavior—worth bookmarking alongside release notes whenever ChatGPT or API defaults shift.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.

Developers gain clearer expectations for how refusal boundaries interact with custom instructions.

Policy teams can map regulatory proposals to explicit value trade-offs instead of opaque RLHF tweaks.

Public specs create a feedback loop: users can cite contradictions when models diverge from stated values.