Any technology which delivers significant value to an organisation, becomes a target for disruption. In recent years AI has emerged as an essential element of cyber defence. Now, the age of AI agents has arrived, and it’s drastically changing the modern threat model. While traditional AI simply made predictions, agentic AI plans, decides and acts. Along the way, it invokes tools and APIs, chains workflows, and interacts with other agents. The inherent autonomy amplifies both value and risk: a single mis‑scoped objective, manipulated prompt, or over‑privileged tool can cascade into data leakage, fraud, codebase drift, or service disruption. And when it happens, it happens at machine speed. How should we prepare to maximise the benefits of AI while minimising the significant risks?