Kubernetes AI Security Playbook: Safeguarding the Data and Model Supply Chain - Andrew Martin & Jack Kelly, ControlPlane

In our quest for AI innovation we often overlook the unique security challenges that come with the AI supply chain: data and model management. Unlike traditional Kubernetes environments AI systems introduce another layer of risk, from data collection to model inference. Attackers can subvert or poison our processes in undetectable ways, and the security of our systems require greater scrutiny than ever. This talk bridges the gap by focusing on the supply chain vulnerabilities unique to AI on Kubernetes. By understanding these nuanced risks, we can better secure our AI assets while harnessing the agility and scalability of cloud native systems. This talk provides actionable insights to enhance security measures in the AI lifecycle on Kubernetes. We:

• Analyze the attack vectors associated with model training data within Kubernetes
• Expose vulnerabilities in a Kubernetes model-training environment
• Illustrate how AI workloads differ from standard Kubernetes deployments in terms of security
• Conduct threat modeling specific to machine learning topologies on Kubernetes
• Propose cloud-native solutions to enhance data integrity and model security