As AI systems move beyond controlled environments into networks, infrastructure, and physical systems, they face adversarial conditions and unpredictable real-world behaviour. Explore what it takes to maintain trust in AI once it operates in the wild, where visibility is limited, threats are active, and outcomes cannot be fully predicted. From cybersecurity to sensing and system resilience, this talk examines how control breaks down outside the lab. When AI leaves the lab, how do we keep trusting what it does?