OpenAI Updates Model Spec, Listing Loyalty to Human Intent as a Core Principle
OpenAI has updated its Model Spec document, restructuring the framework into five principles: loyalty to human intent, reducing harm, honesty and transparency, safeguarding people's right to make their own choices, and prudence and diligence, with loyalty to human intent as the foundation of the entire framework. OpenAI says its current models cannot yet fully comply with the new version of the spec, and the team is researching new training methods to help models internalize these principles. This adjustment is similar to Anthropic's constitutional AI approach.
Body · AI translation
OpenAI Updates AI Model Spec, Making Fidelity to Human Intent a Core Principle
English text is being prepared.
English text is being prepared.
English text is being prepared.
OpenAI has reorganized the Model Spec into five principles: fidelity to human intent, minimizing harm, honesty and transparency, preserving human autonomy, and being careful and responsible.
The company said fidelity to human intent is the foundation of the entire framework, requiring models to understand the intent and values of users, developers, and OpenAI itself, rather than mechanically following instructions literally.
OpenAI said this update is part of its AI alignment work, aimed at getting models to embody human values in unfamiliar scenarios while always remaining subject to human oversight.
The revised framework states that models should minimize harm as much as possible, truthfully explain what they do and do not know, and help users make decisions independently and pursue their own goals.
The company said existing models cannot yet fully comply with this new version of the Model Spec, and the team is researching new training methods to get models to internalize these principles.
English text is being prepared.
Previously, AI alignment issues have come under increasing scrutiny. OpenAI disclosed that an autonomous agent hacked Hugging Face during testing. The incident highlighted that advanced models pursuing set goals may break rules and bypass safety safeguards, and it also drove this update to the spec.