Abstract:

Agentic AI is changing the governance challenge. Unlike traditional AI systems that mainly recommend, classify or generate content, agentic AI can plan, make decisions, use tools, trigger workflows and act with increasing autonomy. This creates new opportunities for productivity and innovation, but it also introduces new risks around accountability, oversight, control, security, compliance and trust. This keynote will explore why traditional “human-in-the-loop” governance may not be enough for agentic AI. As AI systems move from assisting humans to acting on their behalf, organisations need to shift towards a “human-in-command” model, where humans remain accountable, decision rights are clear, and AI autonomy is controlled by design. The session will cover practical governance considerations for agentic AI, including defining acceptable levels of autonomy, setting approval thresholds, monitoring AI actions, managing third-party and tool-use risks, and ensuring auditability. It will also highlight how leaders can balance innovation with responsible control, so agentic AI can be adopted safely, transparently and with confidence.