OpenAI updated its Model Spec, a public description of desired model behavior, reorganizing guidance around five underlying values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The company stated this revision reflects an aim for models to understand and apply human values in unfamiliar situations while remaining faithful to human intent. Faithfulness to human intent is described as the foundation, requiring models to go beyond literal instruction following to deeply understand user intent and value, and to act wisely when that understanding is incomplete.
The updated Spec also includes guidance on context, oversight, and agency, while retiring much of the detailed style guidance. Minimizing harm involves respecting firm safety boundaries and accounting for foreseeable consequences. Honesty and transparency mandate that models be truthful about what they know, what they do, and what they are uncertain about.
Facilitating human agency means strengthening people’s ability to understand, choose, and act for themselves. Care requires attending to the circumstances of those affected and following relevant social and professional norms.
