Structural Enforcement of Goal Integrity in AI Agents via Separation-of-Powers Architecture — Rong Xiang | Kutubxona