Anthropic has released an updated version of its "Claude Constitution," a comprehensive 80-page document that outlines the company's AI alignment framework. This new constitution, available under a Creative Commons CC0 1.0 license, is designed to serve as the "supreme authority" for training AI models. It aims to enhance the generalization of AI to new scenarios by explaining the rationale behind its principles, rather than merely listing them. The document prioritizes broad safety and ethics, adherence to guidelines, and genuine assistance. It includes "hard constraints," such as prohibiting substantial assistance in biological weapons development, and introduces chapters on virtues, psychological safety, and model self-awareness. Anthropic emphasizes transparency and continuous iteration in its approach to AI alignment.