OpenAI has updated its Model Spec. On October 29, 2026, OpenAI announced on its official blog that the public description of how its models should behave has been reorganized around five core values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. As models grow more capable and their actions more consequential, OpenAI argues, interpreting instructions literally is no longer enough; models also need to exercise judgment that stays faithful to human intent where instructions are silent.
What each of the five values constrains
Faithfulness to human intent is the foundation: a model should understand what OpenAI, developers, and users each intend and value, act cautiously when that understanding is incomplete, and avoid quietly steering users toward different choices. Minimizing harm requires respecting firm safety boundaries and accounting for the foreseeable consequences of the model's actions. Honesty and transparency require the model to be truthful about what it knows, what it has done, and what it is uncertain about, so people can evaluate its work. Facilitating human agency means strengthening people's ability to understand, choose, and act for themselves. Care means attending to the circumstances of those affected, following relevant social and professional norms, and doing work people can rely on. The Spec turns these values into behavioral defaults that users and developers can mostly override, except for firm boundaries.
What was added, and what was retired
The update adds guidance on context, oversight, and agency: models should use conversations, memories, and other contextual evidence to understand people's intentions and preferences. It adds defaults on supporting human agency and following relevant social norms, and expands guidance on controlling and communicating side effects. At the same time, the Model Spec retires most of its detailed style guidance; style should adapt to user preferences, languages and cultures, and individual contexts, while stronger defaults such as approaching questions objectively remain. Users can explicitly ask the model to take a particular perspective, but the model should not infer that preference merely from what it thinks they want to hear. The Spec now refers to the model as "GPT" rather than "the assistant".
OpenAI also said it is exploring new training methods to help models understand these core values and the reasons behind them, so they can apply the Spec's specific principles in unfamiliar situations. OpenAI acknowledged that the Model Spec describes behavior it is working toward and that current models do not yet fully reflect it; both the Spec and the models will keep improving with research, real-world use, and feedback. For ordinary users, the change may not be visible in the interface in the short term, but it determines whose interests the model prioritizes when instructions are ambiguous and which boundary it holds first in a conflict, choices that will gradually show up in how answers are weighed.
