OpenAI rewrites its rulebook for ChatGPT around five core values

OpenAI has updated its Model Spec, the document that describes how the company wants its AI to behave. The new version is organized around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. If you use ChatGPT, it is a written statement of what the tool is supposed to do.

What changed

The Model Spec existed before this update. What changed is how it is arranged. OpenAI rebuilt it around the five values instead of leaning on simpler rules. In plain terms, the values are these:

• Faithfulness to human intent: do what the person actually means.
• Harm minimization: avoid causing harm.
• Honesty and transparency: be truthful and open.
• Human agency: keep people in control of their own choices.
• Care: show concern for the person using it.

Why OpenAI did it

OpenAI says its models are getting smarter and taking more consequential actions in the world. As that happens, it says, they need to go beyond following simple rules. Clearer guidance matters more when an AI does bigger things on a person's behalf.

Who it affects

Anyone who uses ChatGPT at home or at work. The Model Spec gives a public standard to measure the tool against. If ChatGPT seems to ignore what you asked, hide something or push you toward a choice, you can compare that behavior with what OpenAI says it intends.

What is not known yet

The full text of the update was not available for this report. It is not yet clear how the new structure will change everyday answers, or how OpenAI will check that its models follow the five values. Readers who want details should read the Model Spec itself on OpenAI's site.