The board approved the system. The system drifted. No alarm sounded. In most organisations, AI governance stops at approval. But AI systems learn, drift, change through vendor updates and increasingly act. Policies, committees and one-time sign-offs may show that governance exists. They do not prove that governance will hold once the system is operating in the real world. The AI Governance Imperative gives boards and leaders a practical governance architecture for AI that learns, changes and acts. It shows how to move beyond policy and approval to an enterprise system that directs value, controls risk, assigns accountability and enables leaders to see whether governance is holding. At its centre is the Behavioural Envelope: a model for defining acceptable AI behaviour, monitoring whether systems remain within agreed limits, producing evidence that controls are working and intervening when those limits are breached. Inside, you will learn how to: Design clear decision rights, escalation pathways and accountability from board to execution - Govern AI portfolios for enterprise value rather than accumulating disconnected pilots - Build lifecycle risk, control, monitoring and assurance into production - Translate regulatory obligations into operational evidence - Manage vendor dependencies, model updates, liability and silent change - Set autonomy boundaries, human oversight and containment for agentic AI The book is supported by 30 original visual models and an end-of-book AI Governance Diagnostic. It is written for boards, executives and leaders across risk, audit, compliance, technology, data, business and transformation. AI is moving from outputs to actions. The question is no longer simply whether your AI works. It is whether you can prove that it is behaving within the limits you set, and act when it is not.