Research
Zombies in Alternate Realities: The Afterlife of Domain Names in DNS Integrations
Many systems outside DNS borrow domain names as identifiers because domains are unique, human-friendly, and already trusted — a TLS certificate is bound to a domain, an ENS name can point a domain at…

- arXiv
- 2605.06880
- Published
- 2026-05-07
- Authors
- Sulyab Thottungal Valapu, John Heidemann, Mattijs Jonker, Raffaele Sommese
AI summary
Overview
Research area: DNS security; the security of systems that link non-DNS namespaces (Web PKI, blockchain naming, software repositories) to domain names.
Technical level: Intermediate. The paper is written for readers who know what DNS, TLS certificates, and blockchain wallets are, but it defines its own terminology (linkage, linked name, linked resource, DNS ownership epoch, zombie linkage) and explains each ecosystem before measuring it.
Scope (1 sentence): The paper defines a threat model for DNS integrations and measures, across Web PKI, ENS, and Maven Central, how often linkages persist after the underlying domain has expired or changed hands — and whether those "zombie" linkages are actively exploitable.
What This Paper Is About
Many systems outside DNS borrow domain names as identifiers because domains are unique, human-friendly, and already trusted — a TLS certificate is bound to a domain, an ENS name can point a domain at a crypto wallet, and a Maven Central package namespace is derived from a domain. Ownership of the domain is checked once, when the linkage is created, but most ecosystems never learn that the domain later expired, was transferred, or was re-registered by someone else. The paper's goal is to define this desynchronization problem as a security issue, measure how prevalent these "zombie linkages" are in three very different ecosystems, and determine which design choices make the problem better or worse.
Key Contributions
- A threat model for DNS integrations. The authors formalize terms (DNS ownership epoch, linkage entry, linked name, linked resource) and define five attack classes: Bulk Linked Name Creation, Linked Name Squatting, Linked Resource Squatting, Linked Name Takeover, and Linked Resource Takeover.
- A method for inferring DNS ownership epochs. Because registries do not publish full registration history, they combine RDAP responses, historical TLD zone delegation data (from dns.coffee), and daily active DNS scans, using an 80-day delegation-gap threshold chosen to match the typical gTLD expiration process.
- Cross-ecosystem measurement of zombie prevalence and persistence. They measure zombies in Web PKI, ENS (both On-chain and Gasless variants), and Maven Central, and relate the differing rates to each ecosystem's design.
- Evidence of live exploitability plus recommendations. They look for zombies that are actively available for abuse in Web PKI and Maven Central, and recommend steps to reduce zombie occurrence.
Main Findings
- Zombies exist everywhere, at very different rates. Roughly 3% of TLS certificates for new domains, 24% of ENS On-chain imports, and 15% of Maven Central namespaces are zombies (the Web PKI figure is given as 2.7% in the body of the paper).
- Validate-once designs accumulate permanent zombies. Of 1,882 active ENS On-chain linkages as of April 2026, 425 (23.8%) are zombies, and none had ever been cleared; ENS On-chain has no revocation mechanism, and the only way to invalidate a linkage is to overwrite it by repeating the process. Of 31,853 Maven Central namespace claims as of March 2026, 4,842 (15.2%) are zombies, and all are permanent under Maven's immutability policy.
- Expiration bounds the damage. Web PKI holds at about 2.7% for newly created domains over the study window (2025-10-01 to 2026-04-15, 52.2 million certificates across 23.6 million newly registered domains, peaking at 35.5 million active certificates on a given day). The authors argue bounded certificate lifetimes are the most effective defense, and that the rarity of revocation (only 4.3% of zombie certificates revoked) shows revocation is unreliable.
- Validating on every use eliminates zombies by design. ENS Gasless linkages are checked against a DNSSEC-signed TXT record on each use and are zombie-free. Since late 2025, over 10k Gasless linkages lost their backing TXT record and became unresolvable — had they been On-chain, every one would have persisted indefinitely.
- Different ecosystems, different causes. ENS On-chain and Maven Central zombies are driven by routine domain expiration. In Web PKI, 39.6% of the 675k zombie-producing domains died within the 5-day Add Grace Period (an upper bound on Bulk Linked Name Creation, since registrar takedowns for abuse could also explain it), with a second mode around 16 days suggesting at least 20% are removed for abuse; nearly 4 in 5 zombie-producing names die within 18 days, leaving most of the certificate's validity as an exploitation window.
- The zombie fraction trends upward where nothing expires. In ENS On-chain the fraction varied from 7% to 30%, with a nadir of 7% in late 2022 during a surge of new registrations, tending upward since 2023. Maven Central's fraction is growing more slowly (15% after 20 years versus ENS's 24% after 7 years), which the authors attribute to software publishers valuing continuity.
- Attacks are actively available, not just theoretical. They find over 192k Web PKI zombie certificates still served months after the DNS name died, 7.3k of which continue to be served past DNS re-registration, and 214 Maven Central zombie namespaces with publishing activity after DNS re-registration.
- Measurement artifacts are acknowledged. The apparent rise and fall of the Web PKI zombie fraction is described as an artifact of new domains needing to die before zombies can form, combined with the 80-day classification threshold; the fraction is stable around 2.7% from about 2025-11 to 2026-02.
- Known biases cut in known directions. Because newly created domains are disproportionately short-lived, 2.7% likely overstates the Web PKI-wide fraction; because only about 25M domains are tracked (while .com alone has over 150 million), the absolute number of zombies ecosystem-wide is likely far larger.
Methodology in Plain English
The approach has four steps. First, the researchers enumerate every linkage in each ecosystem: domain-validated TLS certificates for newly registered domains (found by monitoring Certificate Transparency logs via OpenINTEL's ZoneStream and contacting each name over TLS, keeping only certificates that validate against Mozilla's root store); ENS On-chain linkages (every transaction to the DNSRegistrar smart contract, decoded via Etherscan, using the emitted Claim events); ENS Gasless linkages (TXT records found in OpenINTEL's forward DNS dataset, so the count is a lower bound); and Maven Central namespaces (found by crawling a repository mirror, restricted to reverse-DNS namespaces).
Second, they determine when each linkage was born and died: certificate notBefore and the earlier of notAfter or revocation time (checking CRLs first and falling back to OCSP); the block timestamp of the Claim event, with death defined as the next Claim for the same name; the first published package version for Maven namespaces, which never die.
Third and fourth, they map each linkage's birth to a DNS ownership epoch and ask whether that epoch has ended. Since registry history is not public, they merge three sources — RDAP (authoritative for current registration, but uneven across TLDs and heavily rate-limited), daily TLD zone delegation from dns.coffee (1291 TLDs from 2021, mostly gTLDs in ICANN's CZDS program), and daily NS/SOA scans — into a day-resolution bitset. Runs of observed days become candidate registration intervals; RDAP dates can close or split an interval (splitting signals a re-registration); adjacent intervals are merged only if the gap is under 80 days and RDAP does not object. The authors state their preference for under-counting over over-counting, since falsely labeling a live linkage a zombie would be the more damaging error.
Why This Matters
The paper reframes "the domain expired" from an ordinary DNS event into a cross-ecosystem security problem: a stale linkage can let a previous owner keep publishing software, or let a new domain owner inherit trust they did not earn. It also gives researchers a transferable framework — ownership epochs, linkage lifetimes, revocation, and resource independence — for reasoning about any system that imports DNS names, and it quantifies the outcome of specific design choices rather than treating them as stylistic.
Real-world applications:
- Cryptocurrency users: ENS On-chain zombies let an old wallet mapping survive after the domain changes hands, risking funds sent to the wrong party. ENS Gasless is shown to avoid this entirely.
- Software supply chains: Maven Central namespace immutability means packages published during a previous ownership epoch persist, and 214 namespaces show publishing activity after re-registration — the paper directly extends and quantifies the "MavenGate" risk.
- Certificate authorities and browser/OS root programs: the finding that revocation caught only 4.3% of zombie certificates supports shorter certificate lifetimes as the practical defense.
- Registries and integration designers: the comparison of validate-once, expiring, and validate-on-every-use models gives a concrete menu of mitigations.
Industry relevance: it affects CA/Browser Forum lifetime policy debates, blockchain naming providers choosing between on-chain and gasless designs, and package repository operators such as Maven Central that currently treat namespace claims as permanent.
Future Directions
- Validating the epoch inference against ground truth. The authors state they are obtaining real registration history from a major TLD to check how accurate their inferred epochs are.
- Extending coverage to other integrations. The paper measures three ecosystems and points to others, including Bluesky handles and GitHub Organizations (studied by prior Verisign work that inspired this one).
- Detecting ownership changes that are not re-registrations. The algorithm infers registration epochs, not ownership epochs, so domains sold or transferred without a new registration date — and registrant changes hidden by GDPR redaction and privacy proxies — remain invisible.
- Adopting and testing mitigations. The recommendations in the paper's final section (shorter linkage lifetimes, periodic revalidation, resource independence) are proposed but not evaluated in deployment, and the paper notes that no ecosystem it studied periodically revalidates linkages.
Target Audience
Security researchers working on DNS, Web PKI, blockchain naming, and software supply chains; operators and designers of systems that borrow DNS names as identifiers; certificate authorities and package repository maintainers deciding on expiry and revocation policy; and graduate students looking for a well-structured example of combining heterogeneous public datasets (RDAP, zone files, CT logs, blockchain records) into a measurement study. Readers wanting the specifics of the attack measurements in Section 6 and the ENS/Maven lifespan analysis in Section 5.2 should consult the full paper, since that material is only partly represented in the content summarized here.
Authors’ abstract
DNS integrations leverage the discovery, trust, and uniqueness of the global Domain Name System with a linkage to another naming ecosystem, so the DNS name can help identify resources such as a cryptocurrency wallet or software component. While DNS ownership is verified at linkage creation, many ecosystems do not track subsequent DNS changes. The result is zombie linkages, where the DNS ownership has expired or changed, but the mapping to the linked resource persists. We…
The full abstract on arXiv