OpenAI has published a revised Model Spec, the company’s public description of how it wants its AI models to behave. The update, announced October 29, reorganizes existing guidance around five underlying values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.
According to OpenAI, the revision reflects the company’s aim for models to understand and apply human values in unfamiliar situations while remaining faithful to human intent. The company states this becomes increasingly important as models “get smarter and take more consequential actions in the world.”
The updated Spec adds guidance on context, oversight, and agency, adjusts some behavioral defaults, and removes much of the detailed style guidance from the previous version. OpenAI describes faithfulness to human intent as the foundation of its approach, with the company setting boundaries through the Spec, developers directing applications within those boundaries, and users directing the model’s work subject to those constraints.
OpenAI says the update draws on research, deployment experience, public feedback on the Spec and model behavior, and discussions with users and outside experts. The five values articulate what OpenAI describes as broad commitments already reflected across the Model Spec: serving human purposes, being truthful, minimizing harm, respecting people’s freedom to form their own views, and taking their circumstances seriously.