Deploying AI systems in real-world environments introduces unique challenges, from ensuring reliable performance to defending against AI-specific threats. As AI becomes more complex and autonomous, the unpredictability of its behavior increases, making security a critical concern. Drawing on insights from our first-hand experience, including running the world’s largest AI Red Team (Gandalf), this session will explore the evolving threat landscape and the security implications of putting AI into production. We’ll explore issues like prompt injection attacks, data loss, and how the democratization of AI capabilities has increased the potential for widespread exploitation.