The article exposes the tension between security teams needing visibility into AI workloads and developers who deploy them without formal registration (shadow AI). Traditional security scanners require privileged DaemonSets, kernel-level access, or manual pod-spec edits, which organizations are reluctant to impose because they slow development and compromise cluster stability. This deadlock leaves auditors and SecOps engineers blind to what AI runtimes, models, and dependencies are actually running in production, making compliance with frameworks like the EU AI Act or NIST AI RMF nearly impossible without either breaking developer velocity or accepting unacceptable risk.

To break this deadlock, the authors introduce k8s-aibom, an open-source, unprivileged Kubernetes controller that continuously monitors the cluster API and container environments. It automatically detects AI runtimes (like vLLM and Triton), agent frameworks (LangChain, AutoGen), vector databases (Milvus, Qdrant), and training jobs by inspecting container images, environment variables, and command-line arguments. The controller generates standard CycloneDX 1.6 ML-BOM documents without requiring sidecars, eBPF modules, or developer CI/CD changes. Its deterministic output — identical inputs produce byte-identical BOMs — makes it naturally compatible with GitOps workflows for exact diffs and change-detection alerts. The tool also features a Confidence Model that categorizes assets as Declared, Inferred, or Unresolved, separating clear human intent from autonomous runtime inference to help auditors instantly identify unknowns.

The key takeaway for builders and platform engineers is that k8s-aibom provides audit-grade evidence with least-privilege isolation and immutability. The controller uses minimal IAM permissions and writes ML-BOMs to Google Cloud Storage with DoesNotExist preconditions, making each record cryptographically immutable and tamper-proof. This bridges the gap between low-level Kubernetes runtime state and high-level governance frameworks (EU AI Act, NIST AI RMF, ISO 42001). For teams struggling to move AI projects from pilot to production without compromising security or developer velocity, k8s-aibom offers a zero-friction path that respects both the CISO mandate for visibility and the SRE mandate for stability. It complements existing build-time scanners and cloud posture tools rather than replacing them.