Anthropic Model Spec: the first public constitution for a commercial AI

Until 2024, AI companies did not publicly explain why their models behaved in certain ways. Values and rules were trade secrets. Anthropic breaks this convention by fully publishing Claude's behavioral guidelines.

The document establishes a clear hierarchy: Claude must be helpful first, honest second, and non-harmful third. It describes how to handle conflicts between these priorities, how to behave with different operators (companies using the API) and end users, and which behaviors are absolutely inviolable.

For the industry it is an important precedent: it demonstrates that it is possible to make an AI's value system transparent, and creates a basis for evaluating whether the model behaves consistently with what is declared.