OpenAI has rewritten the rulebook it uses to describe how ChatGPT's models should behave. The company published an updated Model Spec on 29 October, US time, and reorganised the whole thing around five values: faithfulness to human intent, harm minimisation, honesty and transparency, human agency, and care.

File photo of the Pioneer Building in San Francisco, OpenAI’s former headquarters, 2019. Photo: HaeB / Wikimedia Commons (CC BY-SA 4.0), cropped