About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
When the private key behind a website certificate leaks, the certificate stops proving who you are. Anyone holding the key can present your certificate and, with the right network position, pose as your business. The fix has two separate parts: replace the certificate with one made from a brand-new key, and have the old certificate revoked. Replacement is in your hands and should happen the same day. Revocation depends on the certificate authority (CA), the body that issued the certificate, and on how browsers and other software learn about it. You start both at once: you do not wait for the replacement to work before you tell the CA.
The 24-hour figure people quote is a rule for the CA, not for you. The CA/Browser Forum Baseline Requirements say the CA must revoke within 24 hours once it obtains evidence that the key suffered a compromise. Your own duty, in the same document, is to ask promptly and to stop using the key. This guide is for business owners and IT administrators. It explains the two clocks, how revocation information now reaches browsers (OCSP, a live revocation-checking service, is optional for CAs and Let's Encrypt has switched its own service off), and what to settle with a hosting provider before a key ever leaks.
Why it matters
A leaked key is a different problem from an expired certificate. An expired certificate fails loudly and everyone notices. A leaked key keeps working perfectly while it sits in the wrong hands. Nothing in a monitoring dashboard turns red.
The common routes are mundane. A key file copied into a backup that several people can read, a repository that later goes public, a certificate bundle emailed to a contractor. Let's Encrypt, a free CA, lists such routes in its guidance: you might "accidentally share the private key on a public website", or attackers might copy it off a server.
The Baseline Requirements define the term broadly. A key is compromised if "its value has been disclosed to an unauthorized person, or an unauthorized person has had access to it". Our reading: you do not need proof of misuse. Exposure to someone who should not have it is enough to treat the key as compromised.
Who has to act, and how fast
Two different obligations sit in the Baseline Requirements, version 2.3.1 as read for this guide.
The CA's clock. Section 4.9.1.1 says that, with an exception for short-lived certificates, "SHALL revoke a Certificate within 24 hours" applies when the CA "obtains evidence that the Subscriber's Private Key corresponding to the Public Key in the Certificate suffered a Key Compromise". Section 4.9.5 adds that the CA must provide a preliminary report within 24 hours of a certificate problem report, and the period from report to published revocation must not exceed the limits in 4.9.1.1.
Your clock. The "Subscriber" is the certificate holder. Under section 9.6.3, the CA's subscriber agreement must oblige the holder to "promptly request revocation of the Certificate, and cease using it and its associated Private Key" if there is any actual or suspected misuse or compromise of the key. The document gives no number of hours for that duty (our reading of the text: "promptly" is left to judgement).
Put together, our reading is this. The 24 hours begins when the CA has the evidence, which usually means when you tell it. Every hour you spend deciding is an hour the key stays usable. The practical target is hours from discovery to request, not days.
The same section lists other triggers with a longer window. The CA should revoke within 24 hours and must within 5 days for certificates it learns were misused or inaccurate. A key compromise is in the 24-hour group, not the 5-day group.
Anyone can report a compromise. Section 4.9.2 allows relying parties and other third parties to submit certificate problem reports. Let's Encrypt also accepts a revocation request signed with the certificate's own private key, so a finder of a leaked key can revoke on your behalf. Treat that as a reason to move quickly, not as a fix.
Replace and revoke in parallel: a new key, not just a new certificate
Revocation does not undo the exposure. The old key remains in whatever backup, repository history or mailbox it leaked into. Replacement and the revocation request run side by side, and a failed or slow replacement does not pause the request. Our recommendation:
- Tell the CA now. As soon as you suspect the key was exposed, request revocation with the key-compromise reason. Section 9.6.3 asks for this "promptly", and nothing in it makes a working replacement a precondition. Give the CA the date you believe the exposure began.
- Generate a new key pair. Reissuing a certificate with the old key defeats the point, because the exposed key would still be trusted.
- Find every certificate that uses the exposed key. One key can sit behind more than one certificate. Let's Encrypt notes that certificates are logged in Certificate Transparency logs and that searching for a matching SubjectPublicKeyInfo field finds all certificates using a key. Our reading: a public log search is a way to confirm what exists.
- Install the new certificate everywhere the old one runs. That includes load balancers, mail servers, appliances and any service sharing the name.
- Use the key-compromise reason on the request. Let's Encrypt tells subscribers to choose the keyCompromise reason when they believe the key has been exposed to an unauthorised person.
- Clean up the leak itself. Deleting a file from the latest version of a repository does not remove it from history or from backups taken earlier. Treat any copy still readable by someone unauthorised as a continuing exposure (our recommendation).
On order of operations, revoking can take a service offline if the replacement is not ready, and that is a real cost. Leaving the old certificate in use while someone else may hold its key leaves a window of impersonation, and the subscriber duty in section 9.6.3 is to "cease using it and its associated Private Key", not to keep using it until a replacement is convenient. So our recommendation is to decide the outage question early, not to hold the request back:
- Set an escalation point. Agree in advance a time (for example, a few hours from discovery) by which the replacement must be installed. If it is not, the person who owns the decision chooses how to stop using the exposed key: a controlled outage (a maintenance page or taking the affected service offline), or moving the service to a clean replacement certificate and key already available elsewhere. Continuing to serve with the exposed key is not one of the options.
- A controlled outage is acceptable. A short planned outage is better than an open-ended period of serving traffic with a certificate someone else can use.
- Keep the two clocks separate. The CA's 24 hours runs from its evidence; yours is to ask promptly and stop using the key. A late replacement does not move either.
What the CA's revocation record actually does
Revocation is a published statement from the CA that a certificate should no longer be trusted. RFC 5280, the internet standard for certificate and revocation list formats, defines the keyCompromise reason code. It also defines an invalidity date, which "provides the date on which it is known or suspected that the private key was compromised". That date may be earlier than the day the CA processed the revocation. Our reading: if you can say when the key was first exposed, give the CA that date.
Having a revocation on record is not the same as every browser and application refusing the certificate. That depends on how the software finds out, which is where OCSP comes in.
What changed with OCSP
There are two traditional ways software learns about revocation. A certificate revocation list (CRL) is a list of revoked serial numbers the CA publishes. OCSP, the Online Certificate Status Protocol, lets software ask the CA live about one certificate.
OCSP has a privacy cost: the CA learns which site a visitor is opening. One CA, Let's Encrypt, announced a timeline to end its own OCSP service. Its post on 5 December 2024 set 7 May 2025 as the date to stop putting OCSP addresses in certificates and "On this date we will turn off our OCSP responders" for 6 August 2025. It warned that non-browser software such as VPN clients should work with certificates that contain no OCSP address. That is Let's Encrypt's retirement; it does not say every CA has stopped running OCSP, so check which revocation services your own CA still publishes.
The industry rules moved towards making OCSP optional rather than ending it everywhere. The Baseline Requirements revision history records ballot SC063, "Make OCSP optional, require CRLs, and incentivize automation", effective 15 March 2024. Section 4.9.7 requires CAs issuing subscriber certificates to publish a new CRL at least every seven days if all certificates carry an OCSP address, or every four days otherwise, and within 24 hours after recording a revocation.
Browsers do not simply download every CRL. Chromium's documentation says that "Online (i.e. OCSP and CRL) checks are not generally performed by Chrome." Chrome uses CRLSets, which it describes as "the primary means by which Chrome quickly blocks certificates in emergency situations", and which also hold some non-emergency revocations. Mozilla's August 2025 post says Firefox now uses CRLite, which it stores locally and "updates it every 12 hours". It also notes that "Certificate authorities are no longer required to support OCSP".
Our reading, which the sources do not state outright: after a revocation, the time until a given visitor's software rejects the certificate varies by software and update schedule. Different browsers, command-line tools, mobile apps and older embedded software update their lists differently. So a 24-hour CA deadline is a ceiling for publication, not a guarantee that every client has stopped trusting the certificate by then. This is why replacing the certificate and ending use of the old key matter more than waiting for revocation to spread.
Decisions to make before it happens
- Who owns the decision. Name the person who may declare a key compromised and instruct a reissue at any hour. If a provider operates your certificates, the contract should say who can request revocation with the CA.
- Where keys live. List every place a private key is stored or backed up. Keys on hosted servers are often in a backup, and the readers of that backup matter. Revocation does nothing about copies you still hold.
- Who can export. Many servers let administrators mark a key non-exportable. Whether that applies to the platform in use, and who has the rights to override it, is worth knowing before an incident (our recommendation; we did not check platform settings).
- Automation. If certificates renew automatically, a reissue with a fresh key is a small job. If renewals are manual, the fix will take longer than the revocation clock suggests. The shortening maximum lifetimes in section 6.3.2 of the Baseline Requirements (200 days from 15 March 2026, 100 days from 15 March 2027, 47 days from 15 March 2029) will make automation less optional.
- Evidence. Keep the date you learned of the exposure, who was told and when the request was made. You may need it for an incident review.
Read-only checks
These steps look at records; they change nothing. The following command was not tested by the author. It reads a certificate file (the public certificate, never the private key) and prints its serial number, dates, any OCSP address and the names it covers. Options are documented on the OpenSSL x509 manual page.
Linux shell, or any system with OpenSSL:
openssl x509 -in cert.pem -noout -serial -dates -ocsp_uri -ext subjectAltName
Use the serial number and names to look up the certificate in public Certificate Transparency search, and to tell the CA exactly which certificate you mean.
Questions to put to your provider
- Which party holds the private keys for our certificates, and in which backups do copies sit?
- Who can request revocation with the CA, and how do we reach that person out of hours?
- How quickly can the provider generate a new key pair and install a new certificate on every service that uses the old one?
- Are certificates renewed automatically, and does renewal create a new key each time or reuse the old one?
- If an administrator leaves or a backup is lost, is rotating the certificate part of that procedure?
- What do you record, and what can you give us, for an incident review?
A worked example (illustrative)
The following is invented. A firm emails its server certificate and key bundle to an outside developer so a test site can use it. Weeks later it finds the bundle in a shared mailbox that former contractors can still open. The same key protects the main web portal.
The same morning, the administrator asks the CA to revoke the old certificate with the key-compromise reason, giving the date the bundle was sent, and starts the replacement in parallel: a new key and certificate request, then installation on the portal and the mail gateway that shares the name. The firm's own plan sets four hours for the replacement. When the new certificate for the mail gateway fails validation and is still not installed at that point, the owner of the decision takes the gateway's secure listener offline under a controlled outage rather than keep serving the exposed key while the fault is found. The firm also removes the bundle from the mailbox, recalls other copies and records the timeline. The remaining risk is that some devices keep trusting the old certificate until their revocation lists refresh, so the firm checks that nothing still presents it.
Read next
Related articles you may want next.
- HostingWildcard or individual certificates for hosted services: what a shared key puts at riskA wildcard certificate covers the single-label names under a domain with one private key. Where it is reasonable, what it risks, and what to ask your host.Read article
- HostingFinding every certificate before it expires: certificate transparency logs and endpoint readsUse certificate transparency logs and endpoint reads to find forgotten, unknown or unauthorised certificates for your domains before they expire.Read article
- HostingShorter TLS certificate lifetimes: the 2026 to 2029 schedule and what to automateThe CA/Browser Forum schedule cutting TLS certificate lifetimes to 200, 100 and 47 days, and how to decide which renewals to automate first.Read article
Risks and limits
- The Baseline Requirements bind publicly trusted CAs and are enforced through browser root programs. They do not set a legal duty on you. Contract terms with your CA and provider may differ.
- Short-lived certificates are carved out of the 24-hour and 5-day rules in section 4.9.1.1. The Baseline Requirements define a "Short-lived Subscriber Certificate" as one with a validity period of at most 10 days if issued from 15 March 2024 until 14 March 2026, and at most 7 days if issued on or after 15 March 2026. For that defined class, section 4.9.1.1 says only that the CA may support revocation; the deadlines do not apply to it. The Baseline Requirements give no rationale here, and seven days can still be longer than it takes a revocation to reach clients. That is not a licence to leave an ordinary, longer-lived certificate (the usual kind, with up to 200 days from 15 March 2026) in place waiting for it to expire: for those the key-compromise rules above apply. Even for a seven-day certificate, replace the key and stop using the exposed one, because a new certificate for the same key would carry the same exposure.
- We did not verify how individual browsers, operating systems or mobile apps handle revocation today. The Chromium and Mozilla pages describe those two browsers only.
- Private keys also protect more than websites (email, code signing, VPNs). This guide covers publicly trusted TLS server certificates.
- No command here was run, and the article is not lab tested.
Next step
Write down where each private key for your public certificates is stored, backed up and exported, and who is allowed to request a revocation at 2 am. If you are reviewing a hosted service, the Managed Services page is a starting point for that conversation; ask for the scope in writing.
Sources
- Baseline Requirements for the Issuance and Management of Publicly-Trusted TLS Server Certificates, version 2.3.1, CA/Browser Forum, document date 4 October 2026 (published source file, main branch; the same version 2.3.1 text is also on the CA/Browser Forum's own page, read 16 November 2026).
- RFC 5280: Internet X.509 Public Key Infrastructure Certificate and CRL Profile, IETF, May 2008.
- Ending OCSP Support in 2025, Let's Encrypt (Internet Security Research Group), 5 December 2024.
- Revoking Certificates, Let's Encrypt, last updated 31 July 2025.
- CRLSets, The Chromium Projects, date unavailable.
- CRLite: Fast, private, and comprehensive certificate revocation checking in Firefox, Mozilla Hacks, 19 August 2025.
- openssl-x509 manual page (OpenSSL 3.0), OpenSSL Project, date unavailable.
All accessed 15 November 2026.
Learning notes
- A certificate proves identity only while the private key stays private. If the key leaks, the certificate is no longer safe even though it still works.
- Replacing and revoking are different jobs that start together. Replacing is yours; revocation is the CA's action on your request, and a slow replacement never delays the request.
- The 24-hour rule is the CA's deadline from when it has evidence. Your part is to ask promptly and stop using the key, with a controlled outage if the replacement is late.
- OCSP is now optional for CAs, and Let's Encrypt has turned its own service off; browsers increasingly learn about revocation from lists they download, so how quickly they learn varies.
- Reusing the old key in the new certificate defeats the replacement.
This article was originally published on .