AI agents are now production assets with real privileges and real consequences, which means the security conversation has to shift from experimentation to operations. This session takes a practical look at what happens after a Copilot Studio agent goes live and why the biggest risks only become visible once it starts interacting with real users and real data. We dig into the misconfigurations that quietly introduce exposure, the behavioural risks that emerge in real environments, and the visibility gaps that leave security teams guessing. Each of these issues is mapped directly to the OWASP Top 10 Risks for Agentic Applications so you can see how common design and configuration choices line up with recognised threat categories. You will learn where governance and monitoring typically fall short, what telemetry actually reveals about agent behaviour, and the mitigation strategies that make a measurable difference without slowing delivery. This session is built for defenders, platform owners, and makers who want to secure agents as operational workloads and who need a clear view of misconfigurations, risks, and practical controls that hold up in the real world. Doors Open 5.00pm Presentation Starts at 6.00pm Networking 7.00pm Doors Close 7.30pm