Strengthening AI Supply Chain Security on GKE: Introducing k8s-aibom for Automated AI BOMs

'Shadow AI' workloads that developers deploy without separate registration risk bypassing existing security scanners. To address this, Google has open-sourced k8s-aibom.

k8s-aibom is a lightweight, unprivileged Kubernetes controller that continuously monitors the cluster API and container environments. Through this, it automatically detects AI runtimes such as vLLM and Triton and generates a standard CycloneDX ML-BOM (Machine Learning Bill of Materials).

The key features of this tool are as follows:

• Zero Friction: It requires no sidecar injection, kernel-level eBPF modules, or privileged permissions, so it does not compromise cluster stability.
• Real-time Visibility: It generates the BOM based on actual runtime execution state rather than build time, accurately capturing what is currently running.
• GitOps Optimized: It has deterministic characteristics, producing the same BOM for the same input, making it easy to detect changes within GitOps workflows.

Detection Pipeline Stages:

1. Cluster Workload Scraping: Monitors KServe, Deployment, StatefulSet, etc.
2. AI Stack Identification: Analyzes images, environment variables, and command-line arguments to identify AI-related stacks such as vLLM, LangChain, Milvus.
3. Standard Manifest Generation: Compiles the identified artifacts into a CycloneDX 1.6 ML-BOM document.
4. Export to Sink: Attaches the generated BOM to a Custom Resource within the cluster or sends it to destinations such as Google Cloud Storage.