SOC 2, PCI DSS, and HIPAA are different frameworks with different wording, but they ask a small platform team the same five questions. Answering them with evidence rather than policy is the difference between an audit that closes and one that turns into a six-week scramble.

1. Who can get in, and how do you know?

Auditors want to see that access to production is identified, authorised, and reviewed. In Kubernetes that means named identities, no shared cluster-admin kubeconfig, and a record of who held which role and when.

  • Evidence: an export of role bindings and cluster role bindings with subjects and timestamps, plus the join between those identities and your identity provider.
  • The trap: a break-glass cluster-admin credential in a shared password manager that nobody rotates. It will be found, and it will be the finding.

2. How do you control change?

The control is not "we use Git". It is that a change to production can be traced to an approved request, a review, and a deployment, and that nobody can bypass it quietly.

  • Evidence: pull request history with approvals, CI results, and Argo CD or pipeline records showing what was applied, where, and when.
  • The trap: deployments from a laptop. A pipeline that everyone uses except during incidents is a pipeline that will not pass.

3. What do you log, and how long do you keep it?

This is where Kubernetes platforms usually fail first. Audit logging is off by default in many clusters, control plane logs expire, and application logs live in a pod that disappears with the node.

  • Evidence: Kubernetes audit policy and destination, control plane log retention, application log pipeline, and the retention period written down and enforced.
  • The trap: logs retained for three days because the storage bill was too high. The auditor tests the retention policy, not the intention.

4. Is data encrypted, and where are the keys?

In transit and at rest, with a story about key custody that a non-engineer can follow. Managed service encryption counts, as long as nobody claims more than it provides.

  • Evidence: TLS configuration, encryption-at-rest settings per service, and key rotation records.
  • The trap: secrets committed to Git, or encryption keys stored beside the data they protect.

5. Can you recover, and have you proven it?

Backup configuration is easy to produce. A restore that was actually performed, timed, and recorded is the thing that satisfies an auditor and the thing that matters in an incident.

  • Evidence: the backup policy, the most recent restore test with its date and duration, and the runbook for the real event.
  • The trap: a retention policy that has never been tested against a real deletion, and a restore procedure that only exists in someone's head.

The pattern

Every one of those five wants the same thing: a record produced by the platform, not a document describing the platform. Policies are still required, but they are the easy part. The part that takes work is making the system produce the evidence continuously, so that an audit is a matter of exporting rather than reconstructing.

That is also why the work is worth doing beyond the audit. A platform that can answer these five questions is a platform where access is reviewed, changes are reversible, incidents are visible, and recovery has been tested. The certificate is a side effect.

Measured, not guessed.

Tell us what you're dealing with. We'll work out what you need and give you a fixed-price quote. Scoping conversations are free, and we reply within 24 hours.