OpenAI Reorganizes Model Spec Around Five Core Behavior Values
OpenAI published an updated Model Spec on October 29, 2026, grouping its behavioral requirements around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. Most detailed style guidance was retired, and new sections cover context, oversight, and limits on agent autonomy.
- Five values: faithfulness to intent, harm minimization, honesty, agency, care
- Most detailed style guidance retired; style adapts to user preferences
- New instruction hierarchy: Spec, system, developer, then user
- Agent autonomy bounded by ScopeOfAutonomy fields like allowed_tools and max_cost
Read next
Business