Anthropic Writes AI Constitution Governing Claude Behavior

In January, Anthropic published an 80-plus page "AI Constitution" that is trained into its Claude models to govern training, reasoning and behavior. The document embeds a hierarchy of values—prioritizing broad safety, ethical behaviour, compliance, then helpfulness—and instructs Claude to internalize these rules via reinforcement learning and self-critique. The move raises questions about private firms authoring moral frameworks for societally scaled AI.