Domain validation answers exactly one question: did the applicant control this name at the moment the CA checked? That's all it answers. The certificate stays valid until notAfter no matter what happens to the domain afterward. Sell the domain, let it lapse, move off the CDN, cancel the help desk subscription. The certificate has no idea. The private key stays wherever the previous party left it.
We tend to call a certificate "ours" because our hostname is in the SAN list. That's the wrong model. A certificate is a bearer credential issued to whoever held the key on issuance day.
What BygoneSSL showed
Ian Foster and Dylan Ayrey put numbers on this at DEF CON 26 in 2018. Their BygoneSSL research defined a bygone certificate as one issued before the current owner registered the domain, then went looking for them in Certificate Transparency logs. They found roughly 1.5 million domains with certificates older than the current registration, and about 25% of those certificates were still unexpired. Around 7 million more domains shared a certificate with a bygone domain, and 41% of those were still unexpired (their CA/B Forum deck). To make it concrete, they registered insecure.design, got a three-year certificate for it and transferred the domain two months later. Whoever held it first kept a cert valid until 2021.
The shared-certificate number is the one I'd focus on. When a provider packs unrelated customers into one SAN list, any one of those names changing hands gives a stranger standing to get the whole certificate revoked. The researchers called that a denial-of-service vector. Eight years later, it still is.
Those figures are from 2018, when certificate lifetimes were much longer, so don't quote them as current prevalence. The mechanism hasn't changed at all.
What an attacker can actually do with one
Less than the headlines suggest. A bygone cert is not a remote exploit. To impersonate you with it, an attacker has to receive traffic meant for your hostname. That takes a position on the network path, a poisoned resolver, a BGP hijack or a compromised DNS account. And anyone who's in your DNS account can just get a fresh cert anyway.
I still treat these certs as a liability. A valid certificate for your name means a private key sitting in a system you have no contract with and can't audit or rotate. If that vendor gets breached next year, the key leaks and nobody tells you. The existing cert is usually the smaller problem, though. The bigger one is whatever let it be issued, and that's often still sitting in your zone.
Where these certificates come from
Purchased domains get all the attention, but in a typical mid-market estate they're rarely the source. Most bygone certs come from boring vendor turnover:
- CDN and edge migrations. Moving from Cloudflare to Fastly, or Fastly to CloudFront, leaves the old provider holding certs for every hostname it served. Edge providers issue those automatically, so nobody on your team ever touched the key.
- SaaS custom domains: Heroku, Vercel, Shopify, Zendesk help centers, status pages, marketing automation tools. If you CNAME a hostname to a vendor and they handle TLS, there's a cert for your name in their keystore.
- Agencies. An agency builds a campaign microsite on its own hosting. The campaign ends. The CNAME stays, because nobody knows who owns it.
- M&A. Acquired companies bring domains whose certs sit with vendors you've never heard of. Divestitures flip it around: you hand over a domain and keep certs for it, and now you're the bygone party.
- Your own old keys. The cert you replaced on the load balancer last quarter is still valid, and its key may be in an old AMI, a Terraform state file, an etcd backup or someone's home directory. If you rotate without revoking, a working credential stays on disk.
The worse case: delegations that keep renewing
Everything above expires eventually. A setup where you left the vendor a standing way to pass validation never does.
Fastly's TLS setup and Cloudflare's delegated DCV both have you CNAME _acme-challenge for your hostname to a record in their zone. That hands DNS-01 validation to them for as long as the record exists. ACM's DNS validation works the same way. The validation CNAME is tied to the AWS account that requested the cert, and managed renewal checks it again every cycle. So if a vendor requested an ACM cert for your name from their own account, that CNAME lets them keep renewing.
A plain CNAME from the hostname to the vendor does the same thing for HTTP-01, because the vendor still answers for the name. While either record exists, the vendor's automation keeps renewing and the certificate never runs out. I've seen zones with _acme-challenge records pointing at providers the company left years earlier, and those certs renewed the whole time. Nobody reads validation records in a zone with a few hundred entries.
Finding them
Certificate Transparency is what makes any of this findable. Chrome and Safari require CT for publicly trusted certificates, so every cert a browser will accept for your domain is in the logs.
Start with crt.sh. Search your domain with the wildcard prefix to pick up subdomains, and exclude expired entries. Censys certificate search works too, and it's easier to filter by validity and issuer there. The logs contain both precertificates and final certificates, so deduplicate. Then sort what you find:
- Group by issuer. If you buy from Let's Encrypt and DigiCert, a cluster from Google Trust Services or Sectigo probably belongs to a vendor, and the cluster tells you who to contact.
- Compare notBefore against your own dates. For a purchased domain, use the date you took ownership, not the registry creation date, because that doesn't change on transfer. For a vendor, use the cutover date. Certs issued before cutover that are still valid are your bygone set. A cert issued after cutover by the former vendor's CA is worse, because it means a delegation is still live or an old validation got reused. Fix that one first.
- Check the public key. Take the SPKI hash from each CT entry and compare it with the keys you actually hold in ACM, Key Vault, GCP Certificate Manager and your ingress secrets. A key that isn't in your inventory belongs to someone else. This is the check I trust most, because a vendor's cert can have a perfectly normal issuer and date.
- Pull out multi-SAN certs. A cert listing your hostname next to unrelated domains is a shared cert, and it needs a different revocation path.
Getting them revoked
Ask the vendor first. They're the subscriber, and the Baseline Requirements give CAs a 24-hour deadline to act on subscriber requests. In the same message, ask them to delete the custom hostname from their platform, not just stop serving it, or their renewal automation will keep trying. Remove your DNS records pointing at them before they delete it. A deleted hostname with a live CNAME is a subdomain takeover waiting for another tenant to claim it.
If the vendor stalls, revoke it yourself.
ACME revocation
RFC 8555 section 7.6 says an ACME server must accept revocation from three parties: the account that issued the cert, the holder of its private key, or an account with valid authorizations for every identifier in the cert. You want the third. For a Let's Encrypt cert, download it from crt.sh, validate each name with your own account, then run certbot's revoke command against that file. Let's Encrypt's revocation docs cover this case for certificates issued to someone else.
The catch is "every identifier." On a multi-tenant SAN cert you can't validate the other customers' names, so ACME revocation is out.
Certificate problem reports
For everything else, use the CA's problem reporting channel. Get the address from the CCADB, which lists each CA's problem reporting mechanism. The support address on the CA's marketing site is the wrong place. Under BR section 4.9.5, the CA has 24 hours after a report to investigate and send preliminary findings to both you and the subscriber.
Write the report so the CA can act on it without asking you anything. Include the crt.sh ID or serial number, say you're the registrant and offer to prove control by any method they choose. Cite the right revocation reason. BR section 4.9.1.1 requires revocation within five days when "a relevant licensing or services agreement between the Domain Name Registrant and the Applicant has terminated." That describes vendor offboarding almost word for word, and the same clause covers a registrant who failed to renew. The CRL reason code is cessationOfOperation. keyCompromise has a 24-hour deadline, which makes it tempting, but only claim it if you have actual evidence of compromise.
For shared certs, the CA will usually sort it out with the subscriber. The vendor reissues without your name and revokes the old cert as superseded, which is what you wanted anyway.
Don't overestimate what revocation gets you. Chrome's CRLSets cover only a small subset of revocations. Firefox's CRLite covers far more. Revocation moves the cert out of the "valid" column for audits and for clients that check, but the key still exists. That's why offboarding matters more to me than revocation.
Make it part of offboarding
Treat the ability to get certificates for your names like an IAM role. You grant it on purpose and you remove it on purpose. Here's my cutover checklist for any CDN or SaaS vendor:
- Before cutover, export every DNS record that references the vendor: hostname CNAMEs, _acme-challenge CNAMEs, ACM validation CNAMEs, verification TXT records and subzone NS delegations. Pull the certs they've had issued for your names from CT and record the serials.
- Issue your own certs for the hostnames that are moving, then move the traffic.
- Remove every delegation record the same day. Then ask the vendor to revoke and delete the custom hostnames.
- Tighten CAA. Restrict which CAs can issue. Where your CA supports RFC 8657, pin accounturi to your ACME account and limit validationmethods. That's the only CAA control that stops a former vendor that uses the same CA you do, and when half the internet is on Let's Encrypt, that's common. Two caveats. CAA is only checked at issuance, so it does nothing about existing certs. And CAA lookups follow CNAMEs, so a hostname still aliased to the vendor can end up governed by their CAA record instead of yours.
- Alert on CT. A new cert for your names from the former vendor's CA after the cutover date should page someone. So should any cert whose key isn't in your inventory.
What 47-day certificates fix, and what they don't
SC-081v3 helps a lot with the plain bygone case. We're in the 200-day phase right now. A vendor's cert can last up to 200 days, and if their CA reuses validation data, they can get a fresh one up to 200 days after they last validated. In the worst case, one validation is good for over a year of exposure. In March 2029, maximum lifetime drops to 47 days and DCV reuse to 10 days, which brings the worst case under two months.
It does nothing for delegated validation. A vendor with an _acme-challenge CNAME isn't reusing anything. They validate from scratch every time, and shorter lifetimes just mean they do it more often. My bet is that 47-day lifetimes push more vendors to ask for delegation, since it's the obvious way to automate DNS-01 for customer domains. The CA/B Forum's new persistent DNS validation method (BR 3.2.2.4.22, from ballot SC-088v3) works the same way: a standing TXT record that authorizes a specific account until it expires or you delete it. That's useful, but it's also one more record to forget during offboarding.
CertPulse's CT monitoring was built for this problem. It alerts on new issuances for your domains, and its inventory across ACM, Key Vault and GCP Certificate Manager gives you the list of certs you actually hold to compare against. Use it or use something else, but get the alert running before your next vendor migration. If you wait for the incident, you're setting it up too late.
This is why we built CertPulse
CertPulse connects to your AWS, Azure, and GCP accounts, enumerates every certificate, monitors your external endpoints, and watches Certificate Transparency logs. One dashboard for every cert. Alerts when auto-renewal fails. Alerts when certs approach expiry. Alerts when someone issues a cert for your domain that you didn't request.
If you're looking for complete certificate visibility without maintaining scripts, we can get you there in about 5 minutes.