The Good, the Bad, and the Ugly: Hacking 3 Cloud Native AI Services With 1 Vulnerability
Hillai Ben-Sasson & Nir Ohfeld, Wiz
Abstract
With rising demand for AI workloads, SaaS providers are evolving their K8s-based GPU offerings, often built on NVIDIA Container Toolkit: the industry-standard framework for running GPU-based containers. In this talk, we’ll show you how a single vulnerability in this fundamental framework impacted the entire cloud-native ecosystem – and how each environment handled a brand-new 0day.We’ll walk through our discovery of this Pod-to-Node escape vulnerability, and its impact across 3 different providers: Azure, DigitalOcean, and Replicate. Each case began with a weak Pod running our exploit – but outcomes varied widely. One led to minor impact; another with lateral movement that triggered blue teamers; and one with complete service takeover.Join us to gain a firsthand look on how major cloud-native companies build their K8s environments, and the anatomy of a container escape vulnerability in the wild. Finally, learn how to build stronger K8s guardrails by examining the flaws we exploited.
