OpenAI said on October 29, 2026, that it updated its public Model Spec, a document that describes how its GPT models should behave as they take on more consequential tasks.

The revised spec is built around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.

OpenAI said the update also adds guidance on context, oversight, and resolving conflicting instructions, while removing most detailed style guidance.

OpenAI added that its models do not yet fully reflect the spec.

Anthropic has also documented training based on written principles and later introduced an updated public constitution for Claude.

The update comes as the European Union’s AI Act sets criteria for classifying general-purpose AI models as posing systemic risk and requires risk-management systems for high-risk AI systems.

🔗 Source: OpenAI

Recent OpenAI developments