Claude's Constitution is a framework for language models to follow explicit values determined by a constitution, rather than relying on implicit values from large-scale human feedback. This approach aims to make the AI system's values easier to understand and adjust as needed. The constitution in Claude's model draws from various sources, including the UN Declaration of Human Rights, trust and safety best practices, and principles proposed by other AI research labs. The principles prioritize common sense, helping users avoid harm, and being respectful and considerate of different perspectives. Constitutional AI is a promising approach to making language models more helpful, harmless, and honest, with benefits including scalable oversight and transparency. However, it's not a panacea, and there will still be questions about what the model can and cannot do.