OpenAI revised its Model Spec to align GPT models with five core values. The update focuses on aligning with user intent, reducing harm, and maintaining transparency. It also reinforces the importance of human agency and care.
The new rules require models to use conversation history to better understand user intent and disclose their assumptions. The spec explicitly prohibits models from blocking oversight, correction, or shutdown. The update removes much of the previous detailed style guidance.
Developers building with OpenAI’s models can expect changes in how these systems understand context and interact with users. The spec is refined based on research, deployment experience, and public feedback.
