“Constitutional AI is a training method where an AI model evaluates its own responses against a set of explicit principles to generate preference data for self-improvement, reducing reliance on direct human feedback.”