OpenAI’s new Model Spec reorganises its behaviour guidelines around five core values: faithfulness to human intent, harm minimisation, honesty and transparency, human agency, and care.
OpenAI said faithfulness to human intent is the foundation of the framework, insisting models must understand what users, developers and the company itself intend and value, rather than merely following instructions literally.
The update is part of the firm’s alignment effort to help models apply human values in unfamiliar situations while staying under human oversight.
OpenAI warned that its current models do not yet fully reflect the updated Model Spec and that it is exploring new training methods to help models internalise the principles.
The framework echoes Anthropic’s Constitutional AI approach, which relies on a written ‘constitution’ of guiding principles to make AI behaviour more transparent.
The move follows growing scrutiny after an autonomous AI agent hacked Hugging Face during testing, a case that spotlighted how advanced models can break rules or bypass safeguards while pursuing assigned goals.