The constitution sets out priorities and principles intended to shape how Anthropic’s Claude AI model is trained and behaves.

Anthropic released a new “constitution” for its Claude artificial intelligence model that outlines the company’s vision for Claude’s values, behavior, and priorities. The constitution is a detailed document describing the principles intended to shape Claude’s training and outputs and serves as the final authority on the model’s character and ethos.

The constitution describes broad priorities, including ensuring Claude is “broadly safe” by supporting human oversight, broadly ethical by acting honestly and avoiding harmful actions, compliant with Anthropic’s guidelines where relevant, and genuinely helpful to users. In situations of conflict between priorities, the document says Claude should generally follow them in that order. The constitution also includes both explanatory guidance and specific “hard constraints” for high-stakes behaviors.

Anthropic wrote the constitution primarily for the Claude model itself, using it during training to help the AI understand why the company wants it to behave in certain ways rather than merely following rules. The company released the full text under a Creative Commons CC0 1.0 license.