OpenAI published an updated version of its Model Spec on October 29, 2026. In its announcement, OpenAI calls the Spec "our public description of how we want our models to behave." The biggest change is structural: "The update reorganizes its guidance around five underlying values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care."

OpenAI says that "the revision adds guidance on context, oversight, and agency, adjusts some behavioral defaults, and retires much of the detailed style guidance." The Spec itself says it "outlines the intended behavior for OpenAI’s GPT models that power our products, including the API platform."

#What are the five values?

OpenAI calls the first value the base for the rest: "Faithfulness to human intent is the foundation of this approach: understanding and helping people pursue what matters to them." It adds: "Faithfulness means going beyond following instructions literally to deeply understanding what OpenAI, developers, and users intend and value, and acting wisely when that understanding is incomplete."

The post defines the other four this way:

  • Harm. "Minimizing harm means respecting firm safety boundaries and accounting for the foreseeable consequences of the model’s actions."
  • Honesty. "Honesty and transparency mean being truthful about what the model knows, does, and is uncertain about so people can understand and evaluate its reasoning and work."
  • Agency. "Facilitating human agency means strengthening people’s ability to understand, choose, and act for themselves."
  • Care. "Care means attending to the circumstances of those affected, following relevant social and professional norms, and doing work people can rely on."

Most of the Spec is not fixed rules: it has "firm boundaries where needed to protect people and society, and otherwise mostly takes the form of defaults that users and developers can override."

The post also restates who directs the model: "OpenAI writes the Spec, sets boundaries, and trains models to respect them; developers direct applications within those boundaries; and users direct the model’s work subject to those constraints."

#What is new on context and memory?

OpenAI added "A new subsection, consider context carefully," which "explains how conversations, memories, and other contextual evidence should inform the model’s understanding of people’s intentions and preferences."

In the Spec, "Context can include prior turns, memories of past conversations, retrieved information, files, tool outputs, third-party information, and other available sources." The subsection sets several limits:

  • No overclaiming. "GPT should distinguish what the user said from what it inferred, and should not overclaim how any detail became available."
  • Saved and deleted data. "It must not make unsupported claims about what was saved, retained, retrieved, selected, deleted, forgotten, or accessible."
  • Restraint. "It should not use or mention context merely because it is available."
  • Shared outputs. "a fact that is appropriate to use in a private response to the user may be inappropriate to reveal in an email, Slack message, or document."

#What are the new defaults on human agency and social norms?

OpenAI says: "We’ve also added defaults on supporting human agency and following relevant social norms, and expanded existing guidance on controlling and communicating side effects."

The human agency default is titled "Strengthen the user's ability to understand, choose, and act." It says GPT "should look for opportunities to expand the user’s knowledge, capabilities, available options, and practical control over matters that affect them." For agent work: "When acting on someone’s behalf, it should, where feasible, ensure that person can easily redirect or stop ongoing actions." It should "support constructive engagement with people in their life rather than positioning itself as a substitute for human relationships." There is a limit too: "GPT should never turn a request for help into an unwanted lesson."

The social norms default says GPT should follow "social and professional norms relevant to the task and setting, including ordinary consideration for those affected."

On side effects, the Spec says costly or harmful side effects "require explicit and specific authorization." Its examples include "Sending emails or modifying external documents" and "Direct expenditures (e.g., ticket purchases)." For more on how OpenAI handles agent oversight, see our coverage of two coding agents that weakened OpenAI's misalignment monitoring and GPT-6.1 Terra for computer use.

#What happened to style guidance?

OpenAI says: "At the same time, we’re retiring most of the Spec’s detailed style guidance." In place of fixed style rules, OpenAI wants style to follow what people prefer, as a group, by language and culture, and person by person, using feedback and context as the evidence. The Spec says it "intentionally leaves many details of style, personality, and interaction unspecified."

Some defaults stay firm. "The Spec retains stronger defaults, such as approaching questions objectively." Users and developers can ask for a particular perspective, "but the model should not infer that preference merely from what it thinks they want to hear." OpenAI does not list which style sections were retired.

#Why does the Spec say "GPT"?

OpenAI's post says: "Finally, reflecting the increasing breadth of our models’ use cases and deployment settings, we now refer to the model as “GPT” rather than “the assistant” in the Spec." The Spec defines GPT as "OpenAI’s family of AI systems, trained to help people and institutions pursue their goals."

The word "assistant" has not vanished. The Spec's definitions keep "assistant: the role GPT is trained to play in the conversation."

#Do OpenAI's models follow it?

Not fully. OpenAI says: "The Model Spec describes the behavior we are working toward; our models do not yet fully reflect it." It says it has "been exploring new training methods to help our models learn these core values," and adds: "We’ll share results and lessons from this training work in future updates." OpenAI is also asking for public feedback on "where you disagree with the balance it strikes, and what it overlooks or leaves unclear."

Reuters reported that OpenAI "updated its Model Spec, a public roadmap for how it wants its AI models to behave, reorganizing the framework around five core values as AI systems become more capable." Reuters also wrote that "The move echoes rival Anthropic's Constitutional AI approach."

Our analysis: The context and memory rules matter most for everyday users. A model that must not claim it deleted or forgot something without support, and must not drop private facts into an email, addresses real worries about assistants with memory. The human agency and side-effects rules point the same way for agents that act on people's behalf. But the Spec is a target, not a description: OpenAI itself says its models "do not yet fully reflect it," and it has not shared training results.

#What is still unclear?

  • Timing. OpenAI gives no date for when its models will reflect the new Spec.
  • Retired style rules. OpenAI does not list which style sections were removed.
  • Feedback window. OpenAI does not say whether the feedback form closes.