Broken DNSSEC: How to Tell If It Is Taking Your Site Offline, and How to Fix It

When DNSSEC breaks, a domain that is up stops resolving for anyone whose DNS resolver checks signatures, and it keeps working for everyone else. That split is what makes it confusing: your site loads on your phone, a customer in another city gets an error, and nothing on your hosting looks wrong. The usual cause on a small site is a change of DNS provider while the old DS record is still at the registrar. I ran live tests on 19 November 2026 to show what a break looks like from the outside, and checked 71 well-known domains to see how many even use DNSSEC.

Everything below comes from those tests, from Cloudflare’s published account of the .de outage and its DNSSEC troubleshooting guide, or from the protocol rules those pages cite, and I say which is which.

What a DNSSEC break looks like from the outside

DNSSEC adds a signature to your DNS records. A resolver that validates checks the signature against a key, and checks that key against a DS record held by your parent zone (for a .com domain, that is the .com registry, which gets it from your registrar). Cloudflare describes the result as a chain of trust from the root down, and says a break anywhere in the chain makes validation fail for everything below it.

A validating resolver that finds a bad chain does not return your site with a warning. It refuses, and the visitor’s browser reports that the domain cannot be found. A resolver that does not validate never checks, so it still answers. That is why the same domain can work on one network and fail on another.

I used a public test domain whose DS record is deliberately wrong, dnssec-failed.org, and asked both Google (dns.google) and Cloudflare (cloudflare-dns.com) for its A record over DNS-over-HTTPS on 19 November 2026.

Query Google Cloudflare
Normal lookup Status 2 (SERVFAIL), extended error 9: “No DNSKEY matches DS RRs” Status 2 (SERVFAIL), extended error 9: “DNSKEY Missing”
Same lookup with checking disabled (cd=1) Status 0, answer 96.99.227.255 Status 0, answer 96.99.227.255

The pattern in the second row is the test that matters. The address is there and correct. Only validation fails. Cloudflare’s troubleshooting guide uses the same check with dig +cd and says that a good answer with the flag but SERVFAIL without it means DNSSEC is misconfigured. The guide adds that it often happens when authoritative nameservers are changed but the DS records are not updated.

The test domain’s DS record holds a digest that decodes to readable ASCII text, a plea for help. It is a reminder that a DS record is just a string of numbers your registrar passes to the registry, and nobody checks it against your DNS host when you enter it.

Four ways a site owner ends up here

Cause What went wrong Whose fix
Moved DNS provider, old DS left behind The registrar still publishes a DS that matches keys the old provider held. The new provider’s zone is unsigned or signed with different keys. You, at the registrar: remove the DS, or replace it with the new provider’s DS
Wrong DS pasted in A digest, key tag or algorithm number copied with a typo or from the wrong key You, at the registrar: re-enter it from the DNS provider’s own screen
Provider’s key change went wrong Signatures no longer match the keys the zone publishes, or the parent DS no longer matches the new key Your DNS provider, with the registrar if the DS must change
Registry-level failure The TLD operator publishes bad signatures for the whole zone Nobody on your side. Wait, and check the operator’s status

The last row is not hypothetical. Cloudflare reports that on 5 May 2026 at about 19:30 UTC, DENIC, the operator of .de, started publishing incorrect DNSSEC signatures for the .de zone. Validating resolvers, including Cloudflare’s 1.1.1.1, were required to reject them and return SERVFAIL. Cloudflare says the SERVFAIL rate kept climbing for about three hours as cached records expired, and that “serve stale” (RFC 8767) hid part of the damage by answering from expired cache. As a mitigation Cloudflare marked .de as insecure on its own resolver, which has the effect of a Negative Trust Anchor (RFC 7646), an explicit exception that tells a resolver to skip validation for one zone.

Check your own domain in two minutes

You do not need a command line. Paste these into a browser address bar and replace example.com with your domain.

  • Is DNSSEC even on? Open https://dns.google/resolve?name=example.com&type=DS. If the reply has no Answer section, your registrar publishes no DS record, DNSSEC is not on, and DNSSEC cannot be what is breaking your site.
  • Is validation failing? Open https://dns.google/resolve?name=example.com&type=A. A Status of 2 is SERVFAIL. A Status of 0 with an Answer is healthy.
  • Is it DNSSEC specifically? Add &cd=1 to the second address. If the answer appears with the flag and SERVFAIL appears without it, validation is the problem. If both fail, the cause is somewhere else, such as a nameserver that is not answering.

The third check matters. In my survey one domain, hismile.com, returned SERVFAIL both with and without checking disabled and had no DS record, so DNSSEC had nothing to do with it. Do not stop at the first SERVFAIL.

For a visual of where the chain snaps, DNSViz is a free third-party tool that draws it, and Cloudflare’s guide shows an example of an incorrect delegation on it. Verisign also runs a DNSSEC debugger.

The fix, in the order that is safe

  1. Find where the DS record lives. It is set at your registrar, not at your host, unless the same company is both. If you are not sure who your DNS host is, the guide to finding who hosts a website shows how to read the nameservers.
  2. Decide whether you want DNSSEC on the new provider. If you do not, remove the DS record at the registrar. If you do, take the new DS values from the new provider’s DNSSEC screen and replace the old record with them.
  3. Allow for caching. Resolvers keep the DS record for its TTL, so a fix is not instant. The test domain’s DS record carried a TTL of 3,600 seconds (one hour), while Google reported 40,906 seconds for the .com DS record and 19,580 for cloudflare.com on the same morning, so check yours with the DS query above before you decide how long to wait.
  4. Do not delete keys early. Cloudflare’s guide says to wait at least one and a half times the DS TTL after removing the DS record before deleting DNSKEY records, and warns that removing them while DNSSEC is still enabled breaks resolution.
  5. Retest with the three links above from a different network than the one that failed.

When you plan a move, do it in this order: remove the DS, wait for the TTL, change the nameservers. That order means no resolver is ever asked to validate against keys that are no longer served. Cloudflare’s guide names the reverse order, changing the nameservers while the old DS stays, as the usual cause of a break. Email can be hit by the same failure, because the lookup for your mail records is validated like any other, so after a move also confirm that your mail records resolve, as covered in the post on multiple MX records. Whether a sending server bounces or retries when it cannot look you up depends on that server’s settings, which I did not test.

How many sites use DNSSEC at all

I took the 71 domains from my earlier site surveys and asked Google’s resolver, over DNS-over-HTTPS on 19 November 2026, whether each has a DS record at its parent and whether a validated answer comes back.

What I checked Result
Domains with a DS record at the parent 10 of 71
Those 10 that validated cleanly 10 of 10
Domains with a broken DNSSEC chain 0 of 71
.com domains with a DS record 9 of 61
.org domains with a DS record 1 of 6
Signed with algorithm 13 (ECDSA P-256) 9 of the 10 (debian.org uses algorithm 8, RSA/SHA-256)

The ten with DS records are cloudflare.com, debian.org, glossier.com, hostinger.com, hubspot.com, namecheap.com, paypal.com, slack.com, trello.com and yoast.com. The other 61 publish no DS record, so DNSSEC is not protecting them and cannot break them. The .com, .org, .net, .uk and .us zones my domains sit in are themselves signed, and the one .so domain sits in an unsigned zone, so it could not have a DS record at all.

Two readings follow. First, most sites you would copy do not run DNSSEC, so skipping it is ordinary. Second, in my sample none of the ten that do run it was broken on the day. A break needs a change, and a change needs someone to touch the DS record.

My own three sites, veravix.com, repairpicks.com and minisplitsizeguide.com, have no DS record at .com on 19 November 2026, so DNSSEC is off and cannot be the cause of a resolution problem on any of them.

Should you turn it on

DNSSEC protects the integrity of DNS answers. It does not encrypt anything, which Cloudflare states directly, and it does not replace HTTPS. Whether it is worth turning on depends on whether you can keep the DS record in step with your DNS provider every time something changes. If you plan to move DNS hosts, leave it off or write the removal step into the plan before you start.

Sources read on 19 November 2026: Cloudflare’s “When DNSSEC goes wrong: how we responded to the .de TLD outage” and its DNSSEC troubleshooting page, the DNS-over-HTTPS responses of dns.google and cloudflare-dns.com, and the DS lookups described above. The survey of 71 domains ran by script the same morning.