Claude’s constitution makes its intended values visible

Anthropic released a new constitution for Claude, describing principles used to shape the model’s behavior and the reasoning behind them. The document is available under CC0. The company also acknowledges that a model’s actual responses can diverge from the intended principles.

Context

A public statement makes the developer’s choices easier to discuss and scrutinize. It is still a description of the intended behavior, rather than proof that every conversation follows it. The next question is how consistently those principles hold when instructions conflict, context is incomplete, or a conversation becomes difficult.