OpenAI on October 29, 2026 published an updated Model Spec that reorganizes its public behavior rules around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The company says faithfulness to human intent is the foundation. The Spec describes the behavior OpenAI is working toward; current models do not yet fully match it.
Key Facts
| Item | Detail |
|---|---|
| Date | October 29, 2026 |
| Five values | Faithfulness to human intent; harm minimization; honesty and transparency; human agency; care |
| New / expanded guidance | Consider context carefully; oversight; agency; side effects; social norms |
| Retired | Most detailed style guidance |
| Naming in the Spec | “GPT” replaces “the assistant” |
| Spec HTML | https://model-spec.openai.com/2026-10-29.html |
| Changelog | GitHub openai/model_spec, v2026.10.29 |
OpenAI writes that the Spec turns these values into practical defaults, with firm safety boundaries where needed, and otherwise mostly defaults that users and developers can override. It also says it is exploring new training methods so models learn why the principles matter, not only how to follow them literally.
Background
The Model Spec is OpenAI’s public description of how it wants its models to behave across products and the API. The October 29 blog post says the rewrite groups commitments that were already scattered across the Spec: serving human purposes, telling the truth, limiting harm, leaving people free to form their own views, and taking people’s circumstances seriously.
New sections cover faithfulness to human intent (including an updated chain of command and “consider context carefully”), supporting human agency, following relevant social norms, and expanded guidance on controlling and communicating side effects. Style rules are mostly dropped so style can track what people prefer by language, culture, and individual feedback. Stronger defaults such as approaching questions objectively remain.
Reactions and Impact
Reuters, carried on Investing.com and other wires, reported the same five values and that faithfulness to human intent is the base of the update. OpenAI asked readers to say which parts of the Spec they find useful, where they disagree with the trade-offs, and what is unclear.
What We Don't Know Yet
- When production models will substantially match the new Spec
- Which concrete behavior changes users will see first in ChatGPT or the API
- How OpenAI will weigh public feedback against internal research when revising further
- Results from the new value-oriented training methods (promised in later updates)