AI systems face distinct security challenges at different stages of their lifecycle, from build-time (e.g., data collection, training, and validation) to run-time (e.g., deployment and operation). This session explores the evolving threat landscape across these stages, including risks like data poisoning, adversarial attacks, and prompt injection. Attendees will learn how to implement a layered security approach that bridges the gap between build-time and run-time, incorporating practices such as robust model validation, adversarial training, and continuous monitoring. By addressing these challenges holistically, organizations can ensure their AI systems remain secure and resilient throughout their lifecycle.