Training methodology that uses a set of principles to guide AI behavior and reduce harmful outputs, pioneered by Anthropic. Fundamental to Claude's design and safety approach.