An updated Model Spec for more capable and beneficial AI
OpenAI has released an updated version of its Model Spec, the document that sets out how the company wants its models to behave. The revision is framed around five core values that OpenAI says should guide more capable systems toward sound judgment, toward respecting human agency and toward staying faithful to what people actually intend.
The announcement gives no further detail on the individual values, on how they differ from the previous edition, or on how the guidance is applied in training or evaluation. The stated aim is behavioural: as models grow more capable, the spec is meant to keep their decisions aligned with human intent rather than merely compliant with surface instructions. This suggests the emphasis falls on judgment in ambiguous situations, where a model must weigh competing considerations instead of following a single rule.
Why it matters: teams building on OpenAI models, and those writing their own system prompts and safety policies, have a stated reference point for expected behaviour. It is worth reading the full spec before assuming prior guidance still holds, since a stated revision implies changed expectations, though the practical effect on deployed models is not described in this announcement.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
OpenAI’s updated Model Spec outlines five core values to guide more capable AI toward sound judgment, human agency, and faithfulness to human intent.