Explores three key hurdles—generalisation, human behavior, and resources—in translating AI research from lab to real-world, safe deployment.