Discovery Gaps in Internal and Split-Horizon DNS

Discovery Gaps in Internal and Split-Horizon DNS

Split-horizon DNS creates discovery gaps that most subdomain inventories never account for, because the same name can resolve one way for staff on the corporate network and a completely different way, or not at all, for everyone else. This article is for DNS administrators, security engineers and SOC teams who run separate internal and external views, whether through BIND views, Windows Server DNS, Route 53 private hosted zones or Azure Private DNS. It explains where those gaps appear, why external-only subdomain discovery misses them, and how to close them with a repeatable process.

Why Split-Horizon DNS Hides Records From Standard Discovery

Most subdomain discovery works from the outside in. Tools query public resolvers, read certificate transparency logs, check passive DNS datasets and brute-force common labels against your authoritative nameservers. All of that only sees the external view.

Internal views behave differently. A BIND server with a `match-clients` clause for 10.0.0.0/8 serves a separate zone file to internal clients. A Route 53 private hosted zone answers only for associated VPCs. Active Directory-integrated zones answer only on domain controllers. None of these records show up in a public scan, so they never get the same scrutiny.

That leaves two blind spots. Names that exist only internally never get inventoried. Names that exist in both views can drift apart, so the internal copy keeps pointing somewhere the external copy stopped pointing to months ago.

Myth: Internal-Only Records Can’t Be Taken Over

Many teams treat the internal view as safe because it’s unreachable from the internet. That assumption is wrong in a way that matters.

An internal record often points to a public target. For example, a CNAME in the internal zone for `reports.corp.example.com` may point to `contoso-reports.azurewebsites.net`, or an internal alias may point to an S3 website endpoint. The record is private, but the resource it points to is public. If someone deletes that App Service or bucket and an attacker registers the same name, every employee who opens the internal link now lands on attacker-controlled content, while still inside your trusted network perimeter. That’s a dangling DNS record, and external scanners will never report it.

Internal names also leak. Public CAs stopped issuing certificates for non-public names such as `.local` back in 2015, so teams that want valid TLS internally often use ACME DNS-01 validation against a real public subdomain. Every one of those certificates lands in certificate transparency logs. That’s why CT logs as a discovery source regularly expose hostnames like `vault.int.example.com` that the organization considers hidden. VPN clients with split tunnelling can also send internal queries to public resolvers, and those queries end up in passive DNS datasets.

Where Discovery Gaps in Split-Horizon DNS Typically Appear

The most common gap comes from the “same zone name, two servers” setup. If `example.com` is hosted both on Windows DNS internally and on Cloudflare externally, internal clients see only the internal copy. When marketing adds `events.example.com` to Cloudflare, it doesn’t resolve inside the office, so someone adds a hand-made copy to the internal zone. Two years later, the external record has been updated three times and the internal copy still has the original value.

Here’s a realistic case. A company moves its status page from a self-hosted server to a SaaS provider. The external CNAME gets updated on migration day. The internal zone still has an A record pointing to 10.20.4.15, the old VM. Eighteen months later, that IP gets reassigned to a test box running an unpatched admin panel. Staff who type the status URL now reach an unrelated system. No alert fires, because nothing is monitoring the internal view.

Route 53 has its own trap. When a VPC is associated with a private hosted zone for `example.com`, Route 53 doesn’t fall back to the public zone for names that are missing from the private one. It returns NXDOMAIN. Teams fix that by copying public records into the private zone, and those copies then go stale exactly like the Windows example above.

Closing the Gap Step by Step

An experienced DNS administrator starts by treating each view as its own source of truth, not as a variation of the public zone. The process that works in practice looks like this:

1. Export every view on its own schedule. Use `dig AXFR` from an allowed internal host for BIND, `Get-DnsServerResourceRecord -ZoneName` in PowerShell for Windows DNS, and `aws route53 list-resource-record-sets` for each private hosted zone ID.
2. Normalize the exports and diff them by name and type. Any name that exists in both views with different values gets flagged for review.
3. Pull out every internal record whose target is a public hostname, such as `*.azurewebsites.net`, `*.cloudfront.net`, `*.herokuapp.com` or `*.s3-website*`, and check whether the target still exists and is still yours.
4. Check internal A records that point to RFC 1918 addresses against your current IPAM or DHCP data to catch reassigned IPs.
5. Search CT logs and passive DNS for your internal subdomain patterns to see which names have already leaked.

Run this weekly at first. Most teams find 5-15% of internal records need attention in the first pass, then settle on a monthly cadence once the backlog is cleared. Pair this with the external methods covered in discovering all subdomains associated with your domain so both views are covered in one inventory.

Common Mistakes When Auditing Internal Views

The first mistake is auditing from a laptop. An engineer runs `dig` on the VPN, sees an answer and assumes it’s correct. They’ve only tested one view, and depending on resolver order, possibly not the one they meant. Always query each authoritative server directly with `@server`.

The second mistake is forgetting conditional forwarders and stub zones. A forwarder for `partner.example.net` pointing to a partner’s DNS server that was shut down after a contract ended is a dangling delegation in everything but name. If someone takes over that IP, they answer queries for your internal clients.

The third mistake is cleaning up after an incident and stopping there. Teams remove the one stale record that caused trouble but never set up a diff, so the same drift comes back within a quarter.

When the Standard Advice Doesn’t Apply

Air-gapped environments with no public CNAME targets have a much smaller takeover risk, so the focus there should be on reassigned internal IPs rather than cloud resources. Regulated environments, such as banks under DORA or operators subject to NIS2, may restrict zone transfers even internally. In that case, exporting through the management API with a read-only service account is usually easier to justify to auditors than opening AXFR.

Teams under 10 people often don’t need split-horizon DNS at all. A separate internal subdomain like `int.example.com`, delegated to a private resolver, is easier to reason about than two copies of the same zone, and it removes the drift problem completely.

FAQ

Can external subdomain discovery tools find records in my internal DNS view?
Not directly. They only see names that leak through CT logs, passive DNS, public code repositories or misrouted queries. Internal-only records need to be exported from the internal servers themselves.

Is a dangling CNAME in the internal zone really a security risk?
Yes, if the target is a public cloud hostname. An attacker who claims the abandoned resource can serve content to your internal users, who are more likely to trust it because the link came from inside the network.

How often should internal and external views be compared?
Weekly during initial cleanup, then monthly. Also run a comparison after every DNS provider migration, cloud account closure or office network change.

Summary

Split-horizon DNS doesn’t make records safer. It makes them less visible. The internal view needs its own exports, its own diffs and its own checks on public targets, because no external scan will ever cover it. Start with the records that appear in both views but have different values, because they’re both the easiest to find and the most likely to be wrong.