DORA ICT risk requirements for Kubernetes platforms

DORA, Regulation (EU) 2022/2554, requires EU financial entities to run a documented ICT risk management framework. Several of its technical requirements, including access control, network segmentation, encryption and change control, map to Kubernetes settings that policy can enforce and report on. The regulation has applied since 17 January 2025.
What DORA asks of a platform team
Article 6(1) asks for “a sound, comprehensive and well-documented ICT risk management framework” as part of the entity’s wider risk management. Article 6(2) says that framework must include the strategies, policies, procedures and tools that protect information and ICT assets from “damage and unauthorised access or usage”. Article 5(2) makes the management body responsible for approving and overseeing all of it.
Article 6(5) requires the framework to be “documented and reviewed at least once a year”, with a report on the review available to the competent authority on request. Microenterprises get a periodic review instead. Article 4 applies proportionality, so the depth of each control depends on the entity’s size and risk profile.
Platform teams own part of this. They do not own the governance. Board approval, the review cycle and the register of ICT services stay with the financial entity. Article 28(1)(a) says the entity “shall, at all times, remain fully responsible” for its obligations, even when it buys ICT services from a provider.
Which DORA articles map to Kubernetes controls
The table lists the articles where a cluster setting can supply the technical control or the evidence. It leaves out the governance articles.
| DORA article | What it asks for | Kubernetes control | Evidence to keep |
|---|---|---|---|
| Art. 8(1) | Identify and document ICT assets and their dependencies | Namespace, workload and service inventory from the API server | Dated inventory export, reviewed yearly |
| Art. 9(4)(b) | Network infrastructure that can be severed or segmented | NetworkPolicy, with a default-deny policy per namespace | Policy manifests in version control |
| Art. 9(4)(c) | Access limited to what approved functions require | RBAC roles and bindings | RBAC export and access review records |
| Art. 9(4)(d) | Strong authentication and protection of cryptographic keys | Secrets encryption at rest through an EncryptionConfiguration | Encryption configuration and API server flags |
| Art. 9(4)(e) | ICT changes recorded, tested, approved and verified | Admission policy in Audit or Enforce mode | Kyverno policy reports and change tickets |
| Art. 6(5) | Annual review of the framework, reported on request | Dated policy report and audit log exports | Review records linked to the exports |
Each row is a control you still have to design and assign an owner for. A policy shows that a setting is correct at one point in time. It does not show the approval or the test that DORA requires for a change.
Set a Pod Security level for each namespace
- Choose a Pod Security level for each namespace:
privileged,baselineorrestricted. - Set
auditandwarnfirst, so violations show up without blocking deployments. - Set
enforceto the same level once the audit results are clean.
apiVersion: v1
kind: Namespace
metadata:
name: payments
labels:
pod-security.kubernetes.io/enforce: restricted
pod-security.kubernetes.io/audit: restricted
pod-security.kubernetes.io/warn: restricted
The Kubernetes docs define the modes this way. Enforce rejects the Pod. Audit adds an annotation to the audit log event, and warn shows a user-facing warning. Audit and warn also apply to workload resources such as Deployments and Jobs, so teams see violations before a Pod is created. Enforce does not apply to workload resources. Pod Security Admission has been stable since Kubernetes v1.25.
Segment the network with NetworkPolicy
A pod with no NetworkPolicy accepts all traffic. The Kubernetes docs say that “if no policies exist in a namespace, then all ingress and egress traffic is allowed to and from pods in that namespace.” A pod becomes isolated in one direction only when a policy both selects it and lists that direction in policyTypes. Policies are additive, so the order in which they apply does not change the result.
A NetworkPolicy also does nothing without a network plugin that enforces it. The docs say that “creating a NetworkPolicy resource without a controller that implements it will have no effect.” Check this before you rely on any policy.
- Confirm that your network plugin enforces NetworkPolicy.
- Apply a default-deny ingress policy to each namespace.
- Add allow rules for the traffic each workload needs.
- Add a separate egress policy if you need outbound isolation. The default-deny ingress policy does not affect egress.
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: default-deny-ingress
spec:
podSelector: {}
policyTypes:
- Ingress
DORA’s wording on segmentation is specific. The second subparagraph of Article 9(4) says that financial entities “shall design the network connection infrastructure in a way that allows it to be instantaneously severed or segmented in order to minimise and prevent contagion.” A namespace-level default deny is one way to show that segmentation exists. It is not proof that a connection can be cut in an incident, so test it.
Encrypt Secrets at rest, and check the first provider
The Kubernetes docs say that “by default, the API server stores plain-text representations of resources into etcd, with no at-rest encryption.” Encryption depends on how the API server is started, and on the order of providers in its configuration file.
- Check the kube-apiserver flags. Without
--encryption-provider-config, the docs say you do not have encryption at rest enabled. - Check the first provider in the list. If it is
identity, the docs say you still do not have at-rest encryption, because the defaultidentityprovider “does not provide any confidentiality protection.” - Rewrite existing Secrets so they are stored encrypted. Encryption applies to new writes, so old Secrets stay in plain text until they are written again.
# Run as an administrator who can read and write all Secrets
kubectl get secrets --all-namespaces -o json | kubectl replace -f -
The Kubernetes docs describe this command as reading every Secret and updating it with the same data, so that each one is written again through the encryption provider. Run it only after the encryption configuration is active on every control plane host, as the reconfiguration step above requires.
Limit who can change the cluster
Article 9(4)(c) asks for access to be limited to what is required for approved functions. Kubernetes RBAC is where that limit is set. The RBAC good practices page says that “administrators should not use cluster-admin accounts except where specifically needed,” and that a low-privilege account with impersonation rights “can avoid accidental modification of cluster resources.”
Wildcards are the second thing to remove. The page advises: “Avoid providing wildcard permissions when possible, especially to all resources.” A wildcard also grants rights to object types that do not exist yet. Review bindings in each review cycle, and keep the export with the access review records, because DORA asks for the framework to be reviewed as well as set up.
Where policy stops
DORA requires much that no cluster setting produces. Article 24(6) requires that “at least yearly, appropriate tests are conducted on all ICT systems and applications supporting critical or important functions.” Article 28(3) requires a register of information on contractual arrangements for ICT services. Incident reporting and management-body approval are also outside a cluster.
Policy reports answer a narrower question. The Kyverno docs say that reports “always represent the current state of the cluster and do not record historical information.” Keep dated exports of the reports, and pair them with the audit log. The evidence collection guide covers that pairing.
The result values in a policy report carry specific meanings. The table uses the Kyverno docs definitions.
| Result | Meaning in the Kyverno docs |
|---|---|
| pass | The policy evaluated successfully for the resource. |
| fail | The resource did not satisfy the policy. |
| skip | Preconditions were not satisfied (if applicable) in a rule, or an applicable PolicyException exists. |
| warn | A non-compliant result reported as a warning. The annotation policies.kyverno.io/scored has been set to "false" in the policy. |
| error | An evaluation error prevented the policy from producing a normal result. |
Article 6(4), Article 6(6) and Article 24 relax or exclude several requirements for microenterprises. Confirm your entity’s category before you size the programme.
A decision checklist for a platform team
- Confirm that your entity is in scope under Article 2, and which proportionality tier applies under Article 4.
- Map each Article 9 requirement to a named control and an owner.
- Decide, for each control, whether it runs in Audit or Enforce mode.
- Export policy reports and the audit log on a schedule, and keep them with the change records.
- Keep the register of ICT services outside the cluster, with the contract references.
- Review the framework at least yearly, and after any major ICT-related incident, as Article 6(5) requires.
Enforcing this with Kyverno
Kyverno validation rules run in one of two modes. The Kyverno docs say that with Enforce, “resource creation or updates are blocked when the resource does not comply.” With Audit, “a policy violation is logged in a PolicyReport or ClusterPolicyReport,” and the resource is still admitted. Start a DORA control in Audit, check the report, fix the findings, then switch to Enforce.
apiVersion: kyverno.io/v1
kind: ClusterPolicy
metadata:
name: require-ns-purpose-label
spec:
rules:
- name: require-ns-purpose-label
match:
any:
- resources:
kinds:
- Namespace
validate:
failureAction: Enforce
message: 'You must have label `purpose` with value `production` set on all new namespaces.'
pattern:
metadata:
labels:
purpose: production
The example comes from the Kyverno docs. Replace the match and pattern with the control you mapped in the table above. Set failureAction: Audit while you collect results.
For the wider program, read the continuous compliance hub. The SOC 2 controls post, the NIST 800-53 mapping and the NIS2 guide use the same split between enforced checks and evidence. To see the policy engine, start with Kyverno.
Frequently asked questions
Does DORA name specific Kubernetes settings?
No. DORA states outcomes. Article 9 asks for limited access, segmented networks, encryption and controlled change, without naming a tool or a setting. Map each article to a control you can show, and record which evidence supports it. A tool that enforces a control does not make an entity compliant, because the financial entity remains responsible for its obligations under Article 28.
Does DORA apply to a Kubernetes platform vendor?
DORA sets the rules for the financial entities listed in Article 2. A platform vendor is usually an ICT third-party service provider. The financial entity must manage that relationship under Article 28, which includes keeping a register of information on its contractual arrangements. Check your contract and your register entry with your legal and risk teams.
Can Kyverno policy reports serve as DORA evidence?
They help, but not on their own. The Kyverno docs say reports always represent the current state of the cluster and do not record historical information, and that a report is removed when its resource is deleted. Export the reports on a schedule, and keep them with the kube-apiserver audit log and your change records.
What is the difference between Audit and Enforce for a DORA control?
Enforce blocks a create or update that does not comply. Audit admits the resource and logs the violation in a PolicyReport. Run a new control in Audit first to see what it would block, then switch to Enforce after the owners have fixed the findings.