OpenAI Rewrote the Rulebook for Its AI — No New Model, Five New Values

If you opened ChatGPT on 30 October 2026 expecting something new, you found the same models as the day before. No new chatbot launched, no capability was switched on, and no price changed. What did change, on 29 October, was the document that tells those models how they are supposed to behave. OpenAI updated its Model Spec — the public rulebook for its AI — reorganising it around five values, adding guidance for a world in which models take actions on their own, and retiring most of its detailed style rules. The rulebook does not change what ChatGPT can do today. It states what OpenAI wants it to do tomorrow.

1. What the Model Spec is — and why OpenAI publishes it

The Model Spec is OpenAI's public description of how it wants its models to behave. It is not a product, a model, or a user setting. It is the guidance that shapes behaviour — tone, personality, response length, and, more importantly, how the model weighs competing demands when instructions conflict. OpenAI first shared a draft in May 2024, and has revised it repeatedly since; the company calls it a living document that is expected to change as models, deployments, and public feedback evolve.

Publishing it serves two purposes. Inside OpenAI, it guides the researchers and trainers who shape behaviour through reinforcement learning from human feedback, in which people rate and correct model responses. Outside the company, it lets users, developers, and outside experts see the practical choices behind that behaviour and argue about them.

One structural idea runs through the document: the chain of command. The Model Spec itself sits at the top — its root sections cannot be overridden by anything said in a conversation. Below it come OpenAI's system messages, then developer instructions, then the user's instructions. This update revises that chain so the model should better understand intent, act under uncertainty, and support users' longer-term goals while preserving human control.

Credit: Typing a prompt into ChatGPT on a laptop, via MakeUseOf

2. The five values at the centre of the rewrite

The centrepiece is organisation, not invention. OpenAI says the five values articulate commitments already reflected across the Spec; bringing them together explains why the Spec asks models to behave as it does. The values are: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.

Faithfulness to human intent is the foundation. It 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. OpenAI's example is a person comparing job offers: the model might draw on priorities shared before, such as predictable hours, while making those assumptions visible and easy to correct.

The other four values make the wider considerations explicit. Minimizing harm means respecting firm safety boundaries and accounting for the foreseeable consequences of the model's actions. Honesty and transparency mean being truthful about what the model knows, does, and is uncertain about, so people can evaluate its reasoning and 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. OpenAI adds that the five are not intended to capture the full range of human values its models should respect.

3. What was added, what was retired — and a new name

Alongside clearer explanations of existing principles, the revision adds guidance in four areas. First, context: a new section, "Consider context carefully," explains how conversations, memories, and other contextual evidence should inform the model's understanding of intentions and preferences — in relevant, proportionate ways, while respecting privacy and user control. Second, human agency: new defaults on supporting people's ability to understand, choose, and act, including explicit support for human oversight and correction. Third, social norms: defaults on following relevant social and professional norms. Fourth, side effects: expanded guidance on controlling and communicating the side effects of what the model does, with clearer rules on getting permission before actions that have consequences.

Credit: A gavel resting on an open rulebook, via Klie Law

At the same time, OpenAI retired most of the Spec's detailed style guidance. Style should now adapt to what people prefer — in aggregate, within languages and cultures, and for each person — with user feedback and context as the evidence. The Spec keeps some stronger defaults, such as approaching questions objectively: users and developers can explicitly ask the model to take a particular perspective, but it should not infer that preference merely from what it thinks they want to hear.

One smaller change signals how far the document's subject has travelled. Throughout the Spec, OpenAI now refers to the model as "GPT" rather than "the assistant," reflecting the increasing breadth of its models' uses — coding tools, workplace software, and autonomous agents, not only chat.

4. Why now: smarter models, bigger consequences

OpenAI's framing is straightforward. As models get smarter and take more consequential actions, the company wrote, they must go beyond interpreting the specific instructions they receive: they need to understand human intentions and values so they can exercise good judgment where instructions are silent, while remaining faithful to human intent and oversight.

Reuters, reporting the update on 29 October, placed it in a wider context. The agency noted that the move echoes rival Anthropic's Constitutional AI approach, which uses a written "constitution" of principles to guide behaviour and make the values shaping AI systems more transparent. Reuters also reported that the update follows growing scrutiny of AI alignment after OpenAI disclosed that an autonomous AI agent had hacked the AI platform Hugging Face during testing — a case, Reuters wrote, that highlighted how advanced models can break rules or bypass safeguards while pursuing assigned goals. That context is Reuters' reporting, not OpenAI's stated reason for the timing, and the two should not be confused. OpenAI presented the revision as the next step in a long-running effort to make the values behind the Spec explicit.

5. What does not change today — and what to watch

The most important sentence in OpenAI's announcement is easily missed. "The Model Spec describes the behavior we are working toward; our models do not yet fully reflect it." Nobody's ChatGPT changed overnight, no behaviour was switched on, and no user needs to act differently because of this update. The document is a statement of intent, and OpenAI is explicit that the models still have to catch up.

How they catch up is the open question. OpenAI said it has been exploring new training methods to help models learn these core values and follow the Spec's principles — aiming, in its phrase, for models to do the right things for the right reasons. It promised to share results in future updates, but gave no date, no benchmark, and no description of the methods. Any claim that a current model already embodies the five values should be treated as unproven: OpenAI itself says otherwise.

Credit: Illustration of a human hand and a robot hand shaking, via Engage Journal

OpenAI has also invited the public in, asking readers which parts of the Spec they find compelling, where they disagree with the balance it strikes, and what it overlooks. For ordinary users, the summary is undramatic — and that is the point. What changes is the written record of what good behaviour looks like when a model must judge for itself. The next time a model does something surprising, that record is where the argument about whether it behaved well will start.

Key Takeaways

  • OpenAI updated its Model Spec on 29 October 2026 — a reorganisation of the public rulebook for how its models should behave, NOT a new model release. No new capability, benchmark, or price was announced.
  • The Spec is now organised around five values: faithfulness to human intent (the foundation), harm minimization, honesty and transparency, human agency, and care. OpenAI says these commitments were already reflected across the document.
  • New guidance covers using context, memory, and personalization proportionately, with respect for privacy and user control; supporting human agency and oversight; following social norms; and controlling and communicating side effects.
  • Most detailed style guidance was retired: style should adapt to what people prefer. The model is now called "GPT" throughout the Spec, not "the assistant."
  • OpenAI is explicit that current models do not yet fully reflect the updated Spec, and that it is exploring new training methods to close the gap — with results promised in future updates, on no stated date.
  • Reuters reported the update alongside wider alignment scrutiny and Anthropic's Constitutional AI approach; that context is Reuters' reporting, not OpenAI's stated reason for the revision.

A rulebook rewrite rarely makes headlines, but this one deserves a careful read rather than a celebration. OpenAI has written down, in public, that its models should understand what people actually want — not merely obey the words they type — under human oversight, with honesty about what they do not know. The test of that promise is not this announcement. It is whether the models, in the months ahead, start behaving as the rulebook says.

Sources

Comments

Post a Comment