An AI model that performs well on launch day can drift, degrade or fail quietly months later, and in healthcare the consequences fall on patients. This panel looks at what it takes to keep AI safe and trustworthy across its whole life: from procurement and pre-deployment validation, through real-world monitoring and performance drift, to security threats, incident response and knowing when to retire a tool. Panellists will share how their organisations assign responsibility at each stage, what independent assurance looks like in practice, and the guardrails that protect patients and institutions long after go-live.