Kubernetes offers a broad set of security policies, and many Platform Engineering and SRE teams treat a well-configured cluster as a secure one. This whitepaper explains why that confidence is misplaced. Pod Security Admission, NetworkPolicy, RBAC, and admission controllers such as Kyverno and OPA Gatekeeper govern workload configuration, access, and deployment rights well, but they all rely on one unverified assumption: that the scheduled image is exactly what the team built. Policy rules alone leave four questions open: whether an image was altered after its build, where and how it was built, which of its vulnerabilities are actually exploitable, and whether its runtime behavior is legitimate even when it stays within allowed system calls. Admission controllers can check that signatures and attestations exist, but they cannot generate trustworthy evidence that was never created upstream.
The paper sets out a layered trust model in which evidence is produced during the build, and Kubernetes acts as the enforcement point, and it shows how CleanStart puts this model into practice. CleanStart images come from hermetic, reproducible builds using verified source packages, with signed SBOMs and SLSA-aligned provenance generated as part of the build itself. CleanStart SBOM adds reachability and provenance context to that inventory, so admission policies can act on genuinely exploitable risk rather than CVSS thresholds. Runtime activity is then compared against the original build baseline to flag drift. The paper closes with a phased rollout plan and metrics for tracking progress, showing how to build a connected chain from source to runtime in which Kubernetes policies enforce trust that already exists instead of assuming it.