Some Notes on Claude's New Constitution (simonwillison.net)

Anthropic has officially released the "constitution" for its AI model, Claude, as a public domain document under a CC0 license. This follows earlier discoveries by Richard Weiss, who found a substantial internal document that outlined Claude's core values, later dubbed the "soul document." Spanning over 35,000 tokens, this constitution is over ten times longer than the system prompt of Claude Opus 4.5 and serves to clarify the ethical and operational principles guiding the AI's training and usage. The release is significant for the AI/ML community as it provides transparency into the ethical considerations and foundational guidelines informing Claude's development. Notably, the acknowledgments section includes contributions from diverse external reviewers, including members of the Catholic clergy, which suggests a multifaceted approach to ethical review. This initiative reflects a growing emphasis on incorporating broader societal values into AI training processes, potentially influencing future AI governance and ethical frameworks in the industry.