The Good, the Bad, and the Ugly - Hacking 3 Cloud Native AI Services With 1 Vulnerability

Explore a critical security vulnerability in the NVIDIA Container Toolkit through this 32-minute conference talk that demonstrates how a single flaw impacted multiple cloud-native AI services. Discover the speakers' journey in uncovering a Pod-to-Node escape vulnerability and witness its real-world exploitation across three major providers: Azure, DigitalOcean, and Replicate. Learn how each environment responded differently to the same zero-day exploit, with outcomes ranging from minimal impact to complete service takeover and lateral movement that alerted security teams. Gain insights into the architecture and security practices of major cloud-native companies' Kubernetes environments, understand the anatomy of container escape vulnerabilities in production systems, and examine the specific weaknesses that enabled these exploits. Acquire practical knowledge for strengthening Kubernetes security guardrails by analyzing the defensive gaps that were exploited and understanding how the industry-standard framework for GPU-based containers became a vector for widespread compromise across the cloud-native ecosystem.