As cybersecurity and AI converge, the hardest questions are no longer technical but human: AI is entering real decision paths, be it clinical, financial or operational, way faster than our regulatory and privacy models can follow. With agentic AI, systems that plan, act and chain tools on their own, “who reviewed the output?” becomes the far harder question of who is accountable for a sequence of actions no human explicitly approved. During this keynote I draw on hands-on experience as CISO of a Swiss hospital group, where this gap is measured in patient safety — to explore the practical implications of accountability, privacy and machine intelligence, and the questions every leader should ask before an AI agent is given anything that matters.