OpenAI published an update to its public Model Spec on October 29, 2026, reorganizing the document around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The company says the change reflects work from research, deployments, public feedback and expert discussion.
Values first, practical guidance next
OpenAI frames faithfulness to human intent as the foundation of the Spec: models should go beyond following instructions literally to understand users’, developers’ and OpenAI’s values and act accordingly when instructions are incomplete. The update stresses that models should support people’s ability to pursue goals they choose while accounting for effects on others.
The Spec now turns the five values into concrete guidance, using firm boundaries where necessary and otherwise offering defaults that users and developers can override. OpenAI says the Spec retains stronger defaults such as approaching questions objectively, and that style guidance has been largely retired so behavior can adapt to user and cultural preferences.
New sections and behavioral updates
Key additions include an updated chain of command that clarifies which instructions apply and how conflicts are resolved, and a new subsection, “consider context carefully,” on using conversation history, memories and other cues to infer intentions and preferences. The Spec also adds defaults on supporting human agency and following relevant social norms, and expands guidance on controlling and communicating side effects.
OpenAI gives a practical example: when helping someone compare job offers, a model might draw on previously shared priorities—such as predictable hours—while making those assumptions visible and easy to correct. The update also replaces references to “the assistant” with “GPT” to reflect broader deployment contexts.
Alongside the Spec revision, OpenAI says it is exploring new training methods intended to teach models the core values and to help them apply principles in novel situations. The company notes the Model Spec describes the behavior it is working toward and that current models do not yet fully reflect it.
The update was published on October 29, 2026, and OpenAI invites feedback on the values, trade-offs and specific language in the revised Model Spec.
Original source: OpenAI News