Audit Evidence Requirements for DNS Change Records

Audit Evidence Requirements for DNS Change Records

Audit evidence for DNS change records is the part of a SOC 2 or ISO 27001 audit that teams most often underestimate, right up until an auditor picks a random CNAME update from last March and asks who approved it. This article covers what auditors expect to see for DNS changes, which evidence sources hold up under sampling, and how to build a trail that doesn’t require two days of screenshot archaeology every audit cycle.

Why DNS Changes End Up in the Sample

DNS decides where customer traffic and email go. One modified A record or MX record can redirect a login page or a mailbox, so auditors treat DNS configuration as a production system under change management. SOC 2 CC8.1, ISO/IEC 27001:2022 Annex A 8.32 and PCI DSS v4.0 requirement 6.5.1 all cover it.

Sampling usually works the same way. The auditor asks for the full population of DNS changes in the audit period, then picks 10-25 of them. Each one gets traced back to an authorization and forward to a verified result. If the population itself is incomplete, every other piece of evidence becomes questionable.

What Audit Evidence for DNS Changes Must Prove

An experienced compliance lead treats every change record as an answer to five questions:

What changed – record name, type, old value, new value and TTL.
Who made it – a named person, or a service identity that maps to one. A shared “admin” login doesn’t count.
Who approved it – a different person, where segregation of duties applies.
When – provider timestamps in UTC that line up with the ticket.
Verification – proof that live DNS matched the intent after the change.

Before and after values are what’s most often missing. A ticket that says “point app to new load balancer” proves intent. It doesn’t prove what actually changed in the zone.

Evidence Sources That Survive Sampling

Provider-side logs are the foundation. In AWS, every Route 53 ChangeResourceRecordSets call is logged in CloudTrail. Route 53 is a global service, so those events are recorded in us-east-1. That catches out teams whose SIEM only ingests trails from the regions they run workloads in. CloudTrail event history covers 90 days, and anything older needs a trail delivering to S3. Azure’s Activity Log also keeps 90 days by default, so you need to export it to a Log Analytics workspace or storage account. Cloudflare’s audit log records who changed what in a zone, but retention depends on the plan. Check it before audit season, not during.

The retention numbers matter. PCI DSS 10.5.1 requires 12 months of audit log history, with the most recent three months immediately available. SOC 2 Type II periods usually run 6-12 months. Default provider retention almost never covers either one.

The second source is the change system itself: Jira, ServiceNow, or pull requests. When DNS is managed through Terraform or OctoDNS, a merged pull request with a reviewer’s approval and the attached plan output is close to ideal evidence. It shows the diff, the approver and the timestamp in one artifact, which is a big part of why teams move to infrastructure as code for DNS record management.

The third source is independent observation of what the authoritative nameservers actually serve, and it’s the one most teams skip. Scheduled baseline snapshots for detecting unauthorized changes give you a record of the zone over time that doesn’t depend on anyone having logged anything.

A Sampling Exercise That Went Sideways

Picture a SaaS team in its second SOC 2 Type II period with 15 DNS changes sampled. Twelve trace cleanly to merged pull requests. Three don’t.

The first was a TXT record for Google Workspace domain verification. An engineer added it in the Route 53 console at 02:40 on a Saturday during an incident, and no ticket was ever created. The other two were CNAMEs for a marketing vendor, added by a contractor whose IAM user was deleted at offboarding. CloudTrail shows the API calls, but the principal no longer resolves to anyone, and the HR system has no matching record. The result is an exception on CC8.1 and a remediation item in the report.

Both gaps had simple fixes. An emergency change procedure that requires retroactive approval within 1-2 business days would have covered the first. Disabling IAM users instead of deleting them, or federating through an IdP whose sign-in logs outlive the account, would have covered the second.

Myth: The DNS Provider’s Log Is Sufficient Evidence

A provider log proves that an API call happened. It doesn’t prove the call was authorized. It also says nothing about changes made somewhere else: nameserver changes at the registrar, records at a secondary DNS provider, or a subdomain delegated to a vendor’s zone. A registrar-level nameserver change can redirect the entire domain without leaving a single entry in your DNS provider’s audit log.

Completeness is the other half. Auditors increasingly ask how you know the population is complete. The defensible answer is reconciliation: compare logged changes against independent zone snapshots. If a snapshot shows a record that no logged change explains, either logging has a gap or someone has access they shouldn’t.

Common Mistakes When Preparing DNS Evidence

Screenshotting the current zone as proof of a past change. It shows the state today, not the transition on the date being sampled. Auditors spot this quickly, and it reads as reconstructed evidence.

Relying on SOA serial numbers. Route 53, for example, doesn’t increment the SOA serial when records change, so the serial tells you nothing about change history there. Even on platforms that do increment it, the serial shows that something changed, not what or who.

Automation without traceability. A CI pipeline role that writes DNS is fine. A pipeline role whose CloudTrail entries can’t be tied back to a specific commit or pull request isn’t. Pass the commit SHA as a session tag or session name so the log entry points straight at the approved diff.

Building the Evidence Trail Step by Step

Start by listing every place DNS can change: registrars, primary and secondary DNS providers, CDN-managed zones, and delegated subdomains. Then extend logging on each one to at least 13 months. Ship it to immutable storage, such as S3 with Object Lock, or to your SIEM.

Next, route every change through one path, either IaC or a ticket, and limit console write access to a break-glass role that alerts when it’s used. Add daily snapshots of authoritative answers and get an alert on any diff that doesn’t match a known change. Once a quarter, reconcile logs against tickets and snapshots. That reconciled list becomes the population you hand the auditor, and the broader picture of how DNS monitoring supports compliance and audit requirements follows from it.

The standard advice needs adjusting in a few environments. Teams under 10 people often can’t enforce pre-approval segregation, and a documented peer review within 24 hours is a widely accepted compensating control. On-premises BIND shops get the best results from zone files in Git, with dynamic updates restricted to per-person TSIG keys. For Active Directory-integrated DNS, enable “Audit Directory Service Changes” with SACLs on the zone containers so that record edits generate event 5136.

FAQ

How long should DNS change records be retained?
PCI DSS requires 12 months. For SOC 2, keep records for the full audit period plus enough overlap to bridge into the next one. ISO 27001 leaves retention to your own policy. In practice, 13-24 months covers most frameworks.

Do automated ACME validation records need individual approval?
Usually not. Classify _acme-challenge TXT records created by tools like cert-manager or Certbot as pre-approved standard changes in your change policy, and keep the logs that show they were created automatically.

Is DNS in scope for SOC 2 if the provider is a third party?
Yes, if it routes traffic to in-scope systems. The provider’s own SOC 2 report covers their controls, not who in your organization changed your records.

Summary

Strong audit evidence for DNS changes proves what changed, who made the change, who approved it, when it happened, and that the result was verified. Provider logs alone can’t do that, especially for registrar changes or showing the population is complete. Combine extended provider logging, a single approved change path and independent zone snapshots, then reconcile them every quarter. When the auditor asks about last March’s CNAME, the answer is a link, not a weekend of digging.