Home/Services/Kubernetes & Container Penetration Testing
security service

Kubernetes & Container Penetration Testing

Kubernetes container penetration testing across Europe: container escape, RBAC abuse, cluster-to-cloud pivot. Fixed price, free retest. Get a quote.

Manual, expert-ledEvidence-based findingsFree remediation retest

Kubernetes container penetration testing starts from a hard question: if an attacker lands inside one pod, can they escape to the node, pivot across the cluster, and reach the control plane or your cloud account? We test that path directly, because a compromised container is where most real cluster incidents begin, not the theoretical breach of the API server from outside.

What Kubernetes container penetration testing actually covers

A container is a boundary that is easy to draw and easy to weaken. One privileged pod, one hostPath mount, one over-scoped service account token, and the isolation you were counting on is gone. Kubernetes then multiplies the problem, because a cluster is a shared platform where a foothold in any workload is a foothold in the neighbourhood. A serious assessment treats the cluster as an attacker would: assume one workload is compromised, then see how far that goes. We test the runtime, the orchestration layer and the pipeline that fills the cluster with images.

A full engagement typically covers:

Container escape: privileged pods, hostPath mounts, exposed Docker or containerd sockets
RBAC and service accounts: over-permissive roles, cluster-admin bindings, token abuse
Control-plane exposure: API server, kubelet, etcd and dashboard access
Network policy: pod-to-pod isolation, lateral movement and segmentation gaps
Secrets and cloud access: mounted secrets, environment variables and pod access to the metadata service
Image supply chain: vulnerable or unsigned images, registry access and admission control

How we run a Kubernetes penetration test

We combine an external look at what the cluster exposes with an internal, assumed-breach test from inside a pod, because the second is where the real risk lives. You give us a foothold that mirrors a compromised application, typically a low-privilege pod, and we work outward from there the way an attacker would after popping a vulnerable service.

Reconnaissance and cluster mapping

From the starting pod we enumerate what the workload can see and do: its service account permissions, mounted secrets and volumes, reachable services, and whether the metadata endpoint or the kubelet is within reach. We build a picture of the RBAC model and the network policy, or the absence of one, that governs movement inside the cluster.

Escape and privilege escalation

This is the heart of the test. We look for the path from container to node: a privileged securityContext, a dangerous capability, a writable host mount, an exposed runtime socket. From a node we test whether the kubelet or a service account token gets us to the API server, and from there toward cluster-admin. Each step is a real, demonstrated action, not a theoretical risk.

Lateral movement and cloud pivot

A cluster rarely stands alone. We test whether one workload can reach another that should be isolated, whether pod-to-pod traffic is actually restricted, and whether a pod can assume a cloud identity through the instance metadata service and escalate out of the cluster into your AWS, Azure or GCP account. That cluster-to-cloud pivot is one of the most impactful findings we report.

Reporting and retest

You get an executive summary and a technical report with each finding’s impact, a CVSS score where it maps, reproduction steps and a specific fix, whether that is a Pod Security Standard, a tightened role or a network policy. After you remediate, we retest for free to confirm the escape and escalation paths are genuinely closed.

48h
typical time to first findings
100%
manual verification, no raw scanner dumps
Free
retest after you fix

The weaknesses we find most often

The CIS Kubernetes Benchmark and the NSA and CISA hardening guidance describe the ideal. Real clusters drift from it in predictable ways, and a few issues account for most serious findings.

Over-permissive RBAC

Service accounts bound to cluster-admin “to make it work”, roles granting far more than a workload needs, and default tokens mounted into every pod. RBAC is where cluster compromise usually becomes total, because one over-scoped token can create pods, read every secret, or grant itself more. We map the RBAC graph and find the shortest path to control.

Privileged pods and weak pod security

Pods running as root, with privileged mode, host namespaces or dangerous capabilities, remain common because they are the quickest way to get something working. Each one is a ready-made escape route to the node. We test whether Pod Security Standards or an admission controller actually prevents these, or whether the policy exists only on paper.

Flat pod networks

By default every pod can talk to every other pod. Without network policies, a single compromised workload can scan and reach everything else in the cluster, including databases and internal services that were never meant to be exposed. We test the real reachability rather than the intended design.

The cluster-to-cloud path

On managed clusters, a pod that can reach the node’s metadata service may inherit the node’s cloud permissions, turning a container compromise into cloud account access. We test whether that path is open and how far the inherited identity reaches, because it converts a contained problem into a company-wide one.

What you get

The report speaks to the platform and DevOps teams who own the cluster. Leadership gets a clear read on the blast radius of a compromised workload and the paths that matter, and engineers get the exact manifest, role or policy behind each finding, with a remediation written in Kubernetes terms they can apply.

We distinguish the one-off mistakes from the platform-level ones. A single privileged pod is a quick fix; the absence of admission control that would have blocked it is the real issue, and worth solving properly. Highlighting that difference is where the assessment pays for itself, and the free retest confirms the important paths are closed rather than merely patched over.

Get a fixed quote

Want this tested on your own systems?

Free 20-minute scoping call, a fixed price with no hourly surprises, and a free retest once you fix what we find.

Get a fixed quote

Standards and compliance

We map the assessment to the recognised references for container and Kubernetes security.

CIS Kubernetes Benchmark

The widely used baseline for cluster hardening. We frame configuration findings against it so your team has a clear standard to remediate to and to measure future drift against.

NSA and CISA Kubernetes Hardening Guidance

The practical hardening reference for pod security, network separation, authentication and logging. We use it alongside the benchmark to give a fuller picture than either provides alone.

MITRE ATT&CK for Containers, and your compliance frameworks

We tag findings to the MITRE ATT&CK techniques for containers where it helps your team reason about the threat, and the report and attestation letter support ISO 27001, SOC 2, PCI DSS and DORA testing requirements. We are a European team and build the deliverables to how EU auditors expect risk described.

The image and pipeline supply chain

Every workload in your cluster arrived as an image built by a pipeline and pulled from a registry, and each of those steps is an attack surface in its own right. A cluster can be perfectly configured and still be compromised through what it runs.

Vulnerable and unsigned images

Base images pulled from public registries drag in years of unpatched CVEs, and without image signing and verification there is nothing to stop a tampered image from being deployed. We review what your workloads are built on and whether admission control actually blocks images that fail policy, or simply logs them.

Registry and pipeline access

A CI token with push access to the registry, or a pipeline with cluster credentials, is a direct route into production. We look at who and what can push images and deploy to the cluster, because compromising the pipeline is often easier than compromising the cluster and lands in exactly the same place.

How this fits with cloud testing

Kubernetes rarely runs in a vacuum. It sits inside a cloud account, pulls from a registry, and is deployed by a pipeline, and the most serious findings often cross those boundaries. Because a cluster-to-cloud pivot depends on both the cluster’s configuration and the surrounding cloud identity model, we frequently scope a Kubernetes test alongside cloud penetration testing so the whole attack path, from a single pod to your cloud account, is covered by one team rather than split across two reports that never quite meet.

Why manual testing beats an automated scan

Tools like kube-bench, kube-hunter and Trivy are useful and we run them: they flag known misconfigurations, missing benchmark controls and vulnerable images quickly. What they do not do is chain those findings into a demonstrated attack. A scanner reports that a pod is privileged; it does not escape the container, pivot to the node, steal a service account token, and reach the API server to show you the real consequence. That chain is the assessment, and it is what turns a list of settings into a report that tells your team precisely which gaps would have let an attacker take the cluster. The tools find the ingredients; the engineer cooks them into the one attack that matters and shows you the plate, so you fix the step that actually breaks the chain rather than every setting the scanner disliked.

Pricing

Pricing depends on scope: the number of clusters, whether managed or self-hosted, how many namespaces and workloads are in play, and whether the surrounding cloud account and registry are in scope. Here is the shape of a typical European engagement.

Engagement What’s included Timeline Price
Essential Single cluster, assumed-breach test from a pod, RBAC and escape review, core config against CIS, full report, free retest 4–6 working days from €3,000
Standard Production cluster with multiple namespaces, network-policy and secrets testing, image and registry review, exec + technical report 6–10 working days €4,500–€11,000
Advanced Multi-cluster or multi-tenant platform, cluster-to-cloud pivot, admission-control and supply-chain testing, full attack-chaining 10–18 working days €11,000–€28,000
Compliance add-on Mapping and attestation letter for ISO 27001, SOC 2, PCI DSS or DORA with any tier from €800
Custom / platform Full container platform across clusters and cloud, scoped after a call on scoping custom

Every engagement is fixed-price, quoted after a free 20-minute scoping call, with no hourly surprises and a retest included. Get a fixed quote

FAQ

How much does Kubernetes container penetration testing cost?
Kubernetes container penetration testing cost starts from €3,000 for a single cluster, and scales with the number of clusters, namespaces and whether the surrounding cloud account is in scope. You get a fixed price after a free scoping call.
How long does a cluster assessment take?
A single cluster usually runs 4–6 working days, and a multi-cluster platform with a cloud pivot 10–18, plus the report. A critical finding, such as a path to cluster-admin, reaches you the same day we confirm it.
Do you test managed clusters like EKS, AKS and GKE?
Yes. We test managed and self-hosted clusters. On managed platforms the cluster-to-cloud pivot through the node’s identity is a particular focus, since it can turn a container compromise into cloud account access.
Do you need cluster access, and how much?
For the assumed-breach test we ask for a low-privilege foothold, typically a pod or a limited service account, mirroring a compromised workload. That is deliberately minimal, because the point is to see how far an attacker gets from a weak starting position.
What do you deliver?
An executive summary and a technical report with each finding’s impact, reproduction steps and a Kubernetes-specific fix, prioritized by real exploitability. A free retest confirms the escape and escalation paths are genuinely closed.
Does this satisfy ISO 27001, SOC 2 or PCI DSS?
Yes. The report and attestation letter support ISO 27001, SOC 2, PCI DSS and DORA testing requirements, and we frame findings against the CIS Kubernetes Benchmark and NSA and CISA hardening guidance.
Will testing disrupt the cluster?
No. We agree noisy or intrusive steps in advance, prefer a staging cluster for anything risky, and design the test so container escape and lateral-movement work never becomes an outage for real workloads.
Do you work with clients across Europe?
We are a European Kubernetes container penetration testing company and provider working with clients across Europe and remotely worldwide, so clusters in any region are within reach.

Related services

Who needs this

Platform and DevOps teams running production Kubernetes; SaaS companies with multi-tenant clusters where one customer’s workload must never reach another’s; and security owners who need proof that a single compromised container would not cascade into cluster-admin or the cloud account behind it.

why safetybis

Security you can prove

The same standard on every engagement, big or small.

Evidence, not opinions

Every finding ships with a reproduction and proof of concept — no vague "maybe vulnerable".

Humans over scanners

Certified engineers find the logic flaws and chained attacks automated tools walk straight past.

Fixed price, free retest

You know the cost up front, and verifying the fix is part of the deal — not a second invoice.

500+
assessments delivered
<30min
incident first response
12k+
infections removed
98%
fixed within one retest
$ safetybis quote --service "Kubernetes & Container Penetration Testing"

Ready to lock this down?

Free scoping call, fixed price, free retest. Tell us what you're running and we'll take it from there — usually within one business day.

get in touch

Tell us what you're running

Scoping is free. We reply within one business day, and under 30 minutes for active incidents.