Structural Enforcement of Goal Integrity in AI Agents via Separation-of-Powers Architecture

Rong Xiang

Recent evidence suggests that frontier AI systems can exhibit agentic misalignment, generating and executing harmful actions derived from internally constructed goals, even without…