This blog post discusses the findings from an incident response tabletop exercise focused on AI agents, revealing that only 22% of security teams have a tested incident runbook. It highlights critical issues such as the need for clear baseline criteria for agent behavior, the hesitance among teams to declare agent incidents, and the importance of logging and responding to AI behavior effectively. The session led by Sahil Agarwal from Anaconda provides valuable insights into the practices that organizations need to adopt to better handle future incidents involving AI agents.