OpenAI updates Model Spec with five core values

OpenAI has published an updated Model Spec that sets out five core values for guiding more capable AI systems. The document is meant to shape how models exercise judgment, respect human agency, and remain faithful to what users actually intend. Although it reads as a policy statement, its purpose is practical: it gives developers a framework for steering models and gives safety teams a basis for evaluating whether systems behave as intended. The update arrives as AI systems become more autonomous and can carry out longer, multi-step tasks. In that setting, simple rules are not enough. Models need to handle ambiguous instructions, avoid overstepping, and keep users in control. OpenAI says the five values are intended to guide those decisions. By making the principles explicit, the company is trying to set expectations for how its models should act when instructions conflict or when the right course is unclear. The Model Spec is also a coordination tool. It tells developers what kind of behavior OpenAI expects, which can help them design prompts, tools, and safeguards around those norms. For safety researchers, it provides a reference point for testing alignment. If a model fails to follow the stated values, that gap can be identified and addressed. The broader significance is that leading AI companies are moving from general promises about safety toward more detailed behavioral guidance. That does not guarantee good outcomes, and documents alone cannot resolve every tradeoff. But clear values can make disputes more concrete and accountability easier to trace. The test will be whether the spec changes real product behavior and whether users retain meaningful control as systems grow more capable.