OpenAI's October 29 update reorganizes its public Model Spec around five values: faithfulness to human intent, minimizing harm, honesty and transparency, facilitating human agency, and care.

What the update describes

  • The revision groups existing guidance around the five values and adds material on context, oversight, and agency.
  • Some behavioral defaults have been adjusted, and much of the detailed style guidance has been retired.
  • OpenAI says the Spec is a public description of intended model behavior, not a complete account of every human value.

The post frames the update as part of ongoing work informed by research, deployment experience, public feedback, and discussion with outside experts.

For teams building with AI, the public document can inform evaluation criteria, but product-specific behavior still needs testing against the application's own requirements and constraints.