AI governance is moving from boardroom language to security operations, where the question is no longer whether models are powerful, but whether their outputs can be trusted, traced, and defended.