OpenAI has reorganized its Model Spec around human intent, harm reduction, honesty, human agency and care. The public document describes the behavior intended for the company’s models and retains an instruction hierarchy for conflicts. It also addresses oversight, context and user control, while OpenAI acknowledges that production models do not yet fully reflect the specification.
OpenAI has revised Model Spec, its public description of how its models are intended to behave, around five core values. These cover fidelity to human intent, minimizing harm, honesty, human agency and care. The October 29 version retains an instruction hierarchy for resolving conflicts between directions from different authorities. It addresses models used in the company’s products and developer platform.[1], [2]
The human-intent provisions call for supporting oversight and correction, remaining within agreed autonomy limits and treating untrusted data as information rather than instructions by default. The document also directs attention to the origin, scope and freshness of context. Honesty includes being accurate about uncertainty, capabilities and actions taken. Human agency concerns helping people form their own views and make informed choices.[1]
OpenAI acknowledges that models in production do not yet fully reflect the specification. Some style and interaction preferences remain customizable, while stronger safety rules sit higher in the hierarchy. The public document is offered for discussion and reuse under a public-domain dedication. OpenAI presents it alongside usage policies and safety protocols and says feedback and experience continue to inform revisions.[1]
