OpenAI has released an updated Model Spec to guide how its models behave as they take more consequential actions in the world. The update reorganizes guidance around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. This structure reflects OpenAI's work on value alignment, ensuring models understand and apply human values in unfamiliar situations while remaining faithful to intent.
The revision adds guidance on context, oversight, and agency, adjusts behavioral defaults, and retires much of the detailed style guidance. Organizing the Spec around these five values helps explain model behavior and develop clearer guidance where commitments were less fully expressed. OpenAI aims for AI to benefit people with diverse needs and hopes for the future.
Faithfulness to human intent remains the foundation, requiring models to understand what OpenAI, developers, and users intend and value, acting wisely when understanding is incomplete. For example, when comparing job offers, models should draw on shared priorities while making assumptions visible and correctable. The other four values make longer-term interests and effects on others explicit: minimizing harm respects safety boundaries and foreseeable consequences; honesty and transparency model what the model knows and is uncertain about; facilitating human agency strengthens people's ability to understand, choose, and act; and care attends to affected circumstances.
OpenAI writes the Spec, sets boundaries, and trains models; developers direct applications within those boundaries; and users direct model work subject to constraints.
Source: OpenAI Blog · Summarized by HeadlinesBriefing