OpenAI reorganizes its rules for model behaviour around five values
OpenAI says the Spec describes the behaviour it is working toward and that its models do not yet fully reflect it. The update contains no evaluation results showing how closely current models follow it, and no outside party has reviewed whether they do. The training results OpenAI mentions have not been published.
What was claimed?
OpenAI updated its Model Spec, its public description of how it wants its models to behave, on October 29, 2026. The document is now organized around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The update adds a section on how conversations, memories and other context should shape a model's understanding of what people want, adds defaults on supporting human agency and following social norms, expands guidance on controlling and communicating side effects, retires most of the detailed style guidance, and refers to the model as "GPT" rather than "the assistant".
What does the evidence support?
Documented availability, not performance proof
As models take more actions on their own, the rules for what they should do when instructions do not cover a situation matter more. OpenAI says it is also testing new training methods to help models learn these values, and will share results later.
What are the limits?
OpenAI says the Spec describes the behaviour it is working toward and that its models do not yet fully reflect it. The update contains no evaluation results showing how closely current models follow it, and no outside party has reviewed whether they do. The training results OpenAI mentions have not been published.
Claim or evidence date: 2026-10-29. Last verified: .
OpenAI · An updated Model Spec for more capable and beneficial AI ↗
Colors classify evidence; they do not rank truthfulness. Explore hype versus reality →