OpenAI reorganizes Model Spec around five core values

Summary

OpenAI has updated the Model Spec, its public description of how models should behave, around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The revision adds guidance on context, oversight, and side effects, and retires most detailed style rules, keeping the rest as defaults that users and developers can override. The company says it is also testing new training methods to teach these values and will share results in future updates.

Why it matters for our work

As AI agents make more judgment calls in situations no instruction covers, refining those standards in a public document becomes central to practical trust. Organizations using agents can treat the Spec as a reference for out-of-the-box behavior and the boundaries they can tune.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.