DNS root key changes on October 11: which resolvers need checking


APNIC finds uneven readiness ahead of the switch, including a warning signal on Armenia’s Fnet network, but its measurements do not show how many users could be affected.
The DNS root is scheduled to replace the key that signs its public-key list on 11 October 2026. Measurements published on APNIC’s blog on 9 October flag uneven readiness, including on Armenia’s Fnet network, but do not establish how many users could be affected.
The Domain Name System, or DNS, looks up the addresses of websites and other services. A resolver performs those lookups for a device. DNS Security Extensions, or DNSSEC, let a validating resolver check digital signatures to confirm that DNS records are authentic and unchanged. Cloudflare warned in its 6 October explanation that users of a resolver missing the replacement key may be unable to reach websites under any top-level domain, even when the websites are working normally. This is not a forecast of a universal Internet outage, and most website operators need no changes.
This is the second replacement of the root’s key-signing key, or KSK. The first took place in October 2018 after a one-year postponement to give operators more preparation time. KSK-2024, identified by key tag 38696, will replace KSK-2017, key tag 20326, as the signer of the root’s public-key list, called the DNSKEY record set. Validating resolvers must trust the replacement before the switch.
APNIC’s analysis describes two ways to measure whether resolvers trust the replacement key. One counts resolvers that report their trusted keys to root servers. The other uses tests delivered through advertisements to observe the resolver paths used by sampled end users. Their percentages have different denominators and cannot be treated as competing estimates of readiness across the whole Internet.
| Measurement | Reported adoption | Population measured |
|---|---|---|
| Verisign root-server signals, shown as of July 2026 | Around 90% | Reporting resolvers, not weighted by the number of users they serve |
| APNIC advertisement-based sentinel tests | Around 70% | Tests whose resolver paths appear to support the sentinel protocol |
The first method follows RFC 8145, a technical standard under which resolvers put trusted-key tags into queries sent towards root servers. Verisign’s sample counts reporting resolvers, whether each serves one person or millions. APNIC therefore says the figure does not directly estimate the number of users served or potentially affected.
The second method follows RFC 8509, which lets a supporting resolver answer specially named queries about the root keys it trusts. These are called sentinel queries. APNIC’s result is conditional on detecting support for that protocol: it does not mean that 30% of all users are at risk. The share of all users who could be affected remains unknown.
APNIC ran the advertisement-based measurements between May and July 2026 and resumed them in early October. Only a subset of the daily tests clearly showed sentinel support.
| Test group | Daily tests |
|---|---|
| All advertisement-based tests | 16 million |
| Tests showing DNSSEC-validating resolver paths | 6 million |
| Tests clearly signaling sentinel support | 1.8 million |
A separate test using RIPE Atlas network-measurement probes on 5 October covered 2,900 distinct autonomous systems—independently managed networks. APNIC reported the following observations; they describe resolver behavior consistent with a key’s status, rather than direct inspection of resolver configurations.
| Observed resolver behavior | Probes |
|---|---|
| Consistent with sentinel support and KSK-2024 loaded | 1,230 |
| Consistent with KSK-2024 not loaded | 47 |
A supporting DNSSEC-validating resolver first validates the signed address records used by a sentinel query. It then returns the original answer or deliberately substitutes SERVFAIL, a DNS failure response, depending on whether it trusts the queried key. A failure response can therefore be the expected answer, not evidence of a fault.
| Query label | KSK-2024 is trusted | KSK-2024 is not trusted |
|---|---|---|
is-ta-38696 | Valid response | SERVFAIL |
not-ta-38696 | SERVFAIL | Valid response |
APNIC uses three queries. The first asks whether the existing key, 20326, is trusted and expects NOERROR, a successful response. The second asks whether that key is not trusted and expects SERVFAIL. These opposite responses establish sentinel support. The third asks whether replacement key 38696 is trusted: a supporting resolver returns the original answer if it trusts the key and SERVFAIL if it does not.
Cloudflare’s browser readiness test also uses sentinel queries. Its controls check that an ordinary signed name resolves, a deliberately invalid DNSSEC name is rejected, and a sentinel query for the current root key produces a response. If sentinel support cannot be established, the result is inconclusive—not evidence that the replacement key is missing.
The browser test checks only the resolver path used by that browser. Secure DNS settings or a virtual private network (VPN) can affect that path. The result is a snapshot of the resolver answering those requests, not a check of every resolver on a network, and does not replace administrators’ configuration checks.
APNIC lists Armenia’s Fnet, autonomous system AS205368, among 50 networks with low KSK-2024 trust signals. It reports a 38.6% acceptance rate for Fnet in that measurement. This is a network-level test result, not the share of Fnet customers—or Armenian users—who could lose access.
The list includes only networks with at least 1,000 observed tests showing sentinel support. APNIC’s analysis treats the low signals as possible rollover problems warranting checks, not confirmed outages. It recommends that users of listed networks follow up with Cloudflare’s test and ask their provider to review its DNS settings.
DNSSEC checks follow a chain of trust from the DNS root through a top-level domain, such as .com, to an individual domain. The root has no parent domain to confirm its keys, so a resolver starts with a root public key, or its fingerprint, that it already trusts. This starting point is called a trust anchor.
The trusted KSK verifies the root’s public-key list. A separate zone-signing key from that list verifies the root’s other DNS records. Replacing the KSK therefore changes the starting point of the checks, rather than requiring each website to change its own key.
RFC 5011, the standard for automatic trust-anchor updates, lets a resolver authenticate a replacement key using the existing key’s signature on the public-key list. The resolver must observe and verify the replacement, keep checking records containing it for at least 30 days, and verify those records again before accepting it. KSK-2024 has been in the root’s public-key list since 11 January 2025, but each resolver’s waiting period starts with its own first verified observation.
ICANN tells administrators to verify that key tag 38696 is in their resolver’s trust-anchor configuration rather than assume automatic updates worked. If it is missing, its guidance includes checking that automatic updates are enabled and that the resolver can write to its storage directory, then following ICANN’s and the software vendor’s instructions.
Cloudflare says its domain DNS services, 1.1.1.1 resolver and Gateway DNS already trust KSK-2024, so their users need take no action. The company says it added the key to its resolver software’s built-in trust anchors in July 2024. During preparations for the 2018 rollover, some Cloudflare resolvers had lost automatically learned trust-anchor information during software upgrades or moves between machines.
The signing change scheduled for 11 October keeps the same RSA/SHA-256 signature algorithm: it replaces the key pair, not the method used to create and verify signatures. The rollover continues into 2027. According to Cloudflare, ICANN plans to revoke KSK-2017, remove it from the root zone and delete its private key. Ending the old key’s signing role and removing trust in it are separate steps.
Four models lose spoken input commands, but their wired audio connections still work.
2 min

Ninja Forms and WPC Product Bundles users need patched versions and checks for existing compromise: an update alone does not remove attackers’ access.
3 min

Operators are advised to isolate affected devices and restrict access. Exploitation requires local access and existing privileges, rather than a remote attack.
3 min