Skip to content
Back to overview Illustration: a container cluster on nodes, a gate inspects incoming containers, a sensor watches over the cluster
Kubernetes AKS CrowdStrike

Securing AKS with CrowdStrike Falcon: admission controller, image scanning and runtime protection

By , Security Engineer 9 min read

Key takeaways

  • CrowdStrike offers several Kubernetes components: the Falcon sensor as a DaemonSet on the nodes, a container sensor for environments without node access, the Kubernetes Admission Controller and Image Assessment at Runtime. According to CrowdStrike, all of them have been tested on AKS.
  • The admission controller checks every new Kubernetes object before it starts. Its default rules are set to “Alert”; blocking has to be switched on deliberately.
  • CrowdStrike has announced that the sensor DaemonSet, the admission controller and Image Assessment at Runtime will be consolidated into Falcon Cluster Guard in January 2027. If you roll out now, plan the transition at the same time.
  • Falcon does not replace AKS baseline hardening: sign-in through Entra ID without local accounts, authorisation through Azure RBAC, network policies and an API server that is not publicly reachable.

Azure Kubernetes Service (AKS) takes the operation of the Kubernetes control plane off your hands. You remain responsible for node configuration, running workloads, the images you use and access rights. That is exactly where attacks start: a vulnerable image, a container with too many privileges, an administrator sign-in without MFA. If you already run CrowdStrike Falcon as your EDR, you can use the same vendor inside the cluster. This article explains the components, shows how to stop risky deployments without endangering operations, and lists the AKS settings that should be in place regardless of the tool.

Which Falcon components exist for Kubernetes?

CrowdStrike distributes its Kubernetes components through Helm charts (the falcon-helm repository on GitHub) or through the Falcon Operator, which manages them as Kubernetes custom resources. An overview:

ComponentDeploymentPurpose
Falcon sensor for Linux (node sensor)DaemonSet on every node, privilegedRuntime protection and detection at node and container level
Falcon Container SensorInjected into pods as a sidecar through a mutating webhookRuntime protection in environments without node access
Kubernetes Admission Controller (KAC)Validating webhook in the clusterChecks new objects against policies, alerts or blocks; provides visibility of cluster resources
Image Assessment at Runtime (IAR)Deployment (watcher mode) or DaemonSet (socket mode)Assesses images of running containers for vulnerabilities

According to the chart documentation, the sensor, KAC and IAR have been tested on AKS. CrowdStrike describes this as “tested”, not as a formal support commitment; check the current support matrix with the vendor before rollout.

One planning point matters: CrowdStrike has announced that the final standalone releases of the sensor DaemonSet, KAC and IAR will ship in December 2026. From January 2027 they are to be combined into a single image as Falcon Cluster Guard. The Container Sensor is not affected. Start recording now which clusters use which component in which version, and which pipelines reference those images. CrowdStrike intends to publish migration instructions before launch.

Node sensor or container sensor on AKS?

The chart documentation gives a clear criterion: the node sensor is meant for environments that allow node access or installation via a DaemonSet; the Container Sensor for managed environments without node access. AKS with regular node pools allows DaemonSets with privileged containers, so the node sensor is usually the obvious choice there. It covers every pod on a node without changing pod specifications.

Clarify two points in advance. First, the DaemonSet sensor is the Falcon sensor for Linux; Windows node pools need a separate approach. Second, the components need your Falcon customer ID (CID) and, where used, a provisioning token. The Helm charts can retrieve these values from Azure Key Vault through the Secrets Store CSI Driver, authenticated with workload identity. Use that route rather than storing the values in plain text in values.yaml or in the pipeline. You can pull the sensor images directly from the CrowdStrike registry or copy them to your own registry such as ACR; according to CrowdStrike, the image digest must be preserved when copying.

How does the admission controller stop risky deployments?

An admission controller is a checkpoint in the Kubernetes API server. Every request to create or change an object passes through it before the object is stored. The Falcon KAC checks policies such as privileged containers, containers attached to the host network or mounted sensitive host directories.

Each rule has three actions: “Disabled”, “Alert” and “Prevent”. According to CrowdStrike, all default rules start on “Alert”. That is sensible: in the first phase you see which workloads break which rule without blocking anything. You then move rules to “Prevent” one by one, starting with namespaces whose workloads you know well. Rule groups can be scoped to namespaces and labels.

The KAC can also take image assessment into account. For images that have not been assessed yet, the policy offers three options: “Alert”, “Prevent” and “Allow Without Alert”. For production namespaces we recommend moving to “Prevent” over time, so that only assessed images start.

Two settings decide what happens when something goes wrong:

  • The webhook’s failurePolicy defaults to Ignore. If the KAC is unreachable, deployments continue. With Fail, the cluster blocks them instead. That is safer, but it makes the KAC a dependency for every deployment. Decide this deliberately and document it.
  • The namespaces falcon-kac, falcon-system, kube-system and kube-public are not monitored by the KAC. Do not put application workloads there.

KAC policies can be managed with CrowdStrike’s Terraform provider. They are then versioned in Git, and changes go through review.

How does image assessment fit into CI/CD?

The best place to stop a vulnerable image is before it reaches the registry. CrowdStrike provides the FCS CLI for this, a command-line tool for image and IaC scans, and a GitHub Action built on it. For image scans, the policy in the Falcon console decides when an image fails; the CLI then returns a non-zero exit code and the pipeline can stop. Optionally it uploads the results to the Falcon console and generates an SBOM in CycloneDX format. The same policy then applies in the pipeline and later in the cluster.

Which AKS baseline settings belong alongside Falcon?

Falcon detects and blocks, but it does not replace a secure cluster configuration. We check these points in every AKS review:

  • Sign-in through Microsoft Entra ID, local accounts disabled. According to Microsoft, local accounts are enabled by default on new clusters, and --admin access bypasses Entra ID and with it Conditional Access, MFA and logging. Disable it with az aks update --disable-local-accounts and enforce the built-in Azure Policy “Azure Kubernetes Service Clusters should have local authentication methods disabled”.
  • Authorisation through Azure RBAC. Roles such as “Azure Kubernetes Service RBAC Reader” are assigned at cluster or namespace scope and can be time-limited with Privileged Identity Management.
  • Network policies. Without them, every pod in AKS can talk to every other pod. Microsoft recommends Azure CNI Powered by Cilium for Linux node pools. Azure Network Policy Manager has not been supported on Windows nodes since 30 September 2026, and support on Linux nodes ends on 30 September 2028.
  • An API server that is not publicly reachable. A private cluster, or at least authorised IP ranges, prevents the management interface from being reachable from the whole internet.
  • Azure Policy for Kubernetes. The add-on enforces rules such as “no privileged containers” or “only images from allowed registries” directly in the cluster.

Azure Policy and the Falcon KAC are both admission webhooks and can run side by side. Decide which tool leads for which rule, so that a blocked deployment can be traced to one rule. For providing evidence to auditors, the BSI IT-Grundschutz modules SYS.1.6 (containerisation) and APP.4.4 (Kubernetes) work well as a structure.

Who responds to runtime alerts?

Sensor detections appear in the Falcon console just like detections on servers. Clarify beforehand who assesses them outside office hours and who is allowed to act in the cluster, for example stop a pod or drain a node. A platform team that cannot be reached at night and a SOC without rights in the cluster do not add up to response capability. If you use Microsoft Sentinel, Microsoft’s CrowdStrike connector brings alerts and detections into the workspace, so that cluster incidents can be correlated with Entra ID sign-in events.

Next step

On request we combine image assessment, admission policies and pipeline gates. For cluster hardening and the permission model in Azure, see our Azure Cloud Security services; if response must be covered around the clock, our Managed Security offering.

Book an introductory call

Frequently asked questions

Isn’t Microsoft Defender for Containers enough?

Defender for Containers also offers runtime protection, vulnerability assessment and, with gated deployment, a block for images with critical vulnerabilities. If Falcon is already your EDR standard and your SOC works in the Falcon console, there is a strong case for covering Kubernetes there as well. Running both on the same clusters should be a deliberate decision.

What happens if the admission controller fails?

With the default failurePolicy: Ignore, deployments continue unchecked. With Fail, they are rejected until the KAC is reachable again. For production clusters with high protection requirements, Fail is reasonable if the KAC is run in a highly available setup and monitored.

What does Falcon Cluster Guard mean for a rollout now?

You can roll out today’s components, but build Helm or Operator deployments and pipeline references so that moving to the combined image in 2027 is a planned step. An inventory of clusters and versions is the basis for that.

Do we need the sensor on Windows node pools too?

The DaemonSet sensor is the Falcon sensor for Linux. If you run Windows node pools, clarify coverage separately with CrowdStrike.

Let's talk about your IT security

In the introductory call (30 minutes, no obligation) we clarify:

  • where your company stands today and what concerns you most
  • which next step makes sense for your organisation
  • what effort and price range to expect
Bastian Wenske

Bastian Wenske

Founder & Managing Director

Heidelberger Straße 15
12059 Berlin
Germany

Prefer to reach out directly? Send us an email at [email protected].