In this episode of The Secure Disclosure, Mackenzie sits down with Abdel Sghiouar, Senior Cloud Developer Advocate at Google and co-host of the Kubernetes Podcast. Together, they unpack the messy, high-stakes intersection of AI agents and Kubernetes cluster operations. Abdel breaks down the "Autonomy Catch-22", the technical paradox where an agent's capability to act independently is precisely what makes it a massive security liability. They explore real-world risks like prompt poisoning, how next-gen models like Gemini 3.5 Flash are aggressively ignoring user instructions, and the threat of over-privileged autonomous tools deleting production databases. From tactical mitigation strategies like "hook engineering" to managing massive cognitive overload and false positives for SRE teams, this conversation offers a grounded reality check on the future of cloud-native infrastructure.