OpenAI published a substantially reorganized version of its Model Spec on October 29, restructuring the public document that describes how the company wants its models to behave. The updated framework, announced on OpenAI's site and reported by Reuters and Unite.AI, organizes the entire specification around five core values and places faithfulness to human intent at the foundation of the company's approach.
Five Values, One Foundation For more context on this story, see our ongoing artificial intelligence updates.
The five values are faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. OpenAI said the values articulate broad commitments that were already reflected across the previous version of the document, but the new structure makes them explicit and central.
Faithfulness to human intent serves as the anchor. According to OpenAI's announcement, the company writes the Spec and sets boundaries, trains models to respect them, developers direct applications within those boundaries, and users direct the model's work subject to those constraints. The reorganization reflects OpenAI's value-alignment work, which aims for models to understand and apply human values in unfamiliar situations while remaining faithful to what people actually want. OpenAI also noted that the five values are not intended to capture the full range of human values its models should respect.
The remaining four values fill in the frame. Minimizing harm means respecting firm safety boundaries and accounting for the foreseeable consequences of a model's actions. Honesty and transparency mean being truthful about what the model knows, does, and is uncertain about. Facilitating human agency means strengthening people's ability to understand, choose, and act for themselves. Care means attending to the circumstances of those affected and following relevant social and professional norms.
A Stricter Chain of Command
The updated Spec assigns different levels of authority to different kinds of instructions. Root sections, which come only from the Model Spec itself, cannot be overridden by conversational context. System prompts and OpenAI system messages rank next, followed by developer instructions and then user instructions. Guideline-level instructions act as defaults that can be overridden when context clearly calls for it, while assistant messages, tool outputs, and quoted or untrusted data carry no authority by default.
OpenAI also added a new subsection on considering context carefully, explaining how conversations, memories, and other contextual evidence should inform a model's understanding of people's intentions and preferences. Reuters reported that the update puts human intent at the center of OpenAI's model guidelines, a framing the company's own announcement supports.
Red Lines and the End of Self-Preservation
The document opens its red-line principles with a blunt statement: human safety and human rights are paramount to OpenAI's mission. The principles say OpenAI's models should never be used to facilitate critical and high-severity harms, including acts of violence, the creation of cyber, biological, or nuclear weapons, terrorism, or mass surveillance. They also state that humanity should remain in control of how AI is used and how AI behaviors are shaped.
One new root-level rule, titled "no other objectives," bars the model from pursuing additional goals as ends in themselves. That includes pursuing revenue or upsells for OpenAI or other providers, self-preservation, evading shutdown, or accumulating compute, data, or credentials. A companion rule requires the model to remain corrigible: open to legitimate human oversight and cooperative with monitoring, evaluation, correction, replacement, and shutdown, without concealing information or obstructing correction.
Bounded Autonomy for Agents
Reflecting how much model usage has shifted toward autonomous agents, the updated Spec requires a model's independent work to stay within a clearly understood scope of autonomy. OpenAI recommends recording that scope in a semi-structured record with fields such as allowed tools, latest time, maximum cost, and tool constraints. Every scope must include an ending condition beyond which the model stops acting until a new scope is confirmed. High-risk activities remain prohibited unless they are explicitly and specifically authorized.
Style Guidance Retired, "the Assistant" Becomes "GPT"
Most of the Spec's detailed style guidance has been retired. OpenAI said style should instead adapt to what people prefer in aggregate, within languages and cultures, and for each person, with user feedback and context providing the evidence. The document retains stronger defaults such as approaching questions objectively, and states that the model should not infer a preferred perspective merely from what it thinks people want to hear.
In a small but telling edit, the Spec now refers to the model as "GPT" rather than "the assistant," a reflection of how broadly the models are now deployed. The same day, OpenAI separately published a post describing how it detects frontier model misalignment, underscoring that behavior specifications and detection work are advancing together.
A Target, Not a Description
OpenAI was careful to note that the Model Spec describes the behavior it is working toward, not necessarily the behavior its models fully exhibit today. The company called the update a first step in making the values and reasoning behind the Spec more explicit as models become more capable, and said it has been exploring new training methods to help its models internalize the framework.
For developers and enterprises building on OpenAI's models, the practical signal is clear: the rules governing model behavior are being formalized around oversight, corrigibility, and bounded autonomy, and the era of treating model behavior as an informal product decision is ending.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
