Anthropic has released a new constitution for its AI model, Claude. This document outlines the values and behaviors the company wants Claude to embody. It serves as both a guiding framework for Claude’s training and a tool for transparency, helping users understand the intended behaviors of the AI. The constitution is available under a Creative Commons CC0 1.0 Deed, allowing anyone to use it freely. The new constitution replaces a previous version that consisted of standalone principles. The updated approach emphasizes the importance of understanding the reasoning behind certain behaviors rather than just following rules. Claude is expected to act broadly safe, ethical, compliant with Anthropic’s guidelines, and genuinely helpful. In situations where these priorities conflict, Claude should prioritize them in that order. The constitution also includes specific sections discussing helpfulness, compliance with guidelines, and ethical behavior, with detailed explanations on navigating complex situations. For example, it underscores the importance of Claude being a reliable source of information while recognizing the need for human oversight during AI development. The document reflects Anthropic’s evolving strategy to train AI models like Claude. By using the constitution during various stages of training, Claude can generate synthetic data that aligns with its values, enhancing its ability to make sound judgments in real-world scenarios. This approach aims to foster good behavior and ethical decision-making, even as the challenges of developing powerful AI models continue to grow.