The eMudhra CA Key Compromise: What Companies Should Check and Fix Now

On Saturday 17 October 2026 a security researcher reported that eight intermediate certificate authorities run by eMudhra, an Indian certificate authority, had private keys anyone could work out from the public certificates. Every one of the keys was a number smaller than 256. One of them was the number 4. Six of the eight could issue TLS certificates trusted by Apple, Google, Microsoft and Mozilla; the other two could sign Windows software.
Chrome blocked the six TLS keys an hour and forty minutes after the report, and Firefox the same evening. For people browsing the web, the danger was over by Saturday night.
It was not over for everything else. Servers, containers, Java applications, Python scripts, payment integrations and connected devices check certificates against their own copy of the list of trusted authorities, and most of those copies change only when someone installs an update. Until then, a certificate for any domain signed with one of these keys passes as genuine, and the keys have been public since 20 October.
In the same week GoDaddy found 403,643 certificates whose policy information did not match its own published documents, and revoked every one that had not expired first. Its customers had to download and install the replacements themselves.
This guide is for the people who have to answer "are we affected?": IT and operations leads, CTOs, and founders whose business runs on a website, an API or a connected product. It covers what happened, taken from the incident records Mozilla, Google, eMudhra, GoDaddy and the software vendors have published; what to check today; what to update this week; and what to put in place so the next incident of this kind is routine. Everything here is as of 13 November 2026, and the incident is still open.
What Happened
eMudhra created the eight intermediates at a key ceremony on 30 April 2025. Its full incident report, filed in Mozilla's incident bug on 10 November 2026, explains what went wrong. The hardware security module generated the keys correctly. They then went through a wrap and unwrap step, approved as a deviation from eMudhra's usual procedure, to prepare a move to another vendor's modules, using a Python script its operations team had written for the job. After the unwrap, the script read each private key out of the module into the host's memory and created a new key object from it, and for the eight ECDSA keys "a single line in that read-back truncated the private value to its leading byte". Those truncated keys, not the ones the module had generated, were used to create the certificate requests, and eMudhra's root signed them. The same requests went to Sectigo, which cross-signed all eight from its AAA Certificate Services root on 14 May 2025. Nothing in the ceremony tested the strength of the keys. Eight RSA intermediates from the same ceremony kept their full keys, but those keys had passed through the host's memory too, so eMudhra revoked them on 7 November.
Mozilla described the result: "P384 private keys are randomly selected integers between 1 and the group order. The 8 impacted keys all used integers less than 256." Mozilla received the report at 20:06 UTC and recovered the keys itself at 21:48, "without any knowledge of the method the original reporter used". The researcher says he found them with Google's open source paranoid_crypto library, "out of the box". The weak keys had been part of publicly trusted chains for almost 18 months.
The eight certificates:
| Name | Use |
|---|---|
EM DV TLS CA - G3B, EM OV TLS CA - G3B, EM EV TLS CA - G3B | TLS (eMudhra BV, Netherlands) |
EM DV TLS CA - G3C, EM OV TLS CA - G3C, EM EV TLS CA - G3C | TLS (eMudhra Inc, United States) |
EM OV CS CA - G3B, EM OV CS CA - G3C | Code signing |
Each key sits in two certificates. One was issued by eMudhra's emSign Root TLS CA - G3 or emSign Root CS CA - G3, which chain up to emSign ECC Root CA - G3; the other is Sectigo's cross-certificate from AAA Certificate Services. The second path matters, because some trust stores that never contained an eMudhra root still trust that old Sectigo root.
| When (UTC) | What |
|---|---|
| 17 October, 18:05 | The researcher reports to Apple, Google, Microsoft, Mozilla and eMudhra |
| 17 October, 19:45 | Chrome blocks the six TLS keys through a CRLSet update |
| 17 October, 22:59 | Firefox blocks them through OneCRL |
| 18 October, 09:24 to 09:38 | eMudhra revokes all eight |
| 19 October | Sectigo revokes its eight cross-certificates. Apple begins blocking the keys on its devices. Android receives the block through Google Play services around 22:00 |
| 20 October | Public announcement. The Windows disallowed certificate list blocks all eight keys. Mozilla removes emSign ECC Root CA - G3 from its trust store (NSS 3.131.1). Updates for Firefox, Go, Debian, Ubuntu, Alpine, Rust, Electron and Cloudflare's clients |
| 21 October | Updates for Node.js, Deno and FreeBSD; OpenJDK and Oracle advisory CVE-2026-91403; Red Hat bulletin without a fixed package |
| 22 October | eMudhra announces that it will leave the TLS certificate business. Fixed packages for Red Hat Enterprise Linux 9 and 10 |
| 23 October | Apple also blocks emSign ECC Root CA - G3. Fixed packages for Red Hat Enterprise Linux 7 (ELS) and 8 |
| 24 October | OpenJDK replaces its advisory with CVE-2026-114820, which removes four roots from Java's own trust store. Updates for Amazon Linux and Corretto |
| 26 October | Microsoft publishes a statement. The root list that Windows downloads no longer contains emSign ECC Root CA - G3 or AAA Certificate Services |
| 27 October | Chrome's removal of emSign ECC Root CA - G3 from its root store (version 43) is scheduled to reach all devices in the morning, US Pacific time. Fixed packages for Red Hat Enterprise Linux 6 (ELS) and OpenShift |
| 29 October | eMudhra publishes its answers to Chrome's questions |
| 2 November | Fix for the Amazon Linux 2027 Preview. eMudhra postpones its full incident report to 9 November |
| 4 and 5 November | Google and Mozilla set a cut-off for eMudhra's other root, emSign Root CA - G1: TLS certificates first logged in CT after 16 November will not be trusted. Temurin 8 and 11 updated; Ubuntu fixes Java's keystore |
| 7 November | eMudhra revokes the eight RSA intermediates created at the same ceremony |
| 9 November | macOS 27.2.1, 26.7.3 and 15.8.3 fix the command-line tools. InCommon starts issuing DV certificates through SSL.com |
| 10 November | eMudhra files its full incident report. Microsoft announces that it will remove the remaining ten eMudhra roots, emSign Root CA - G1 among them, from the Windows and Edge root lists on 16 November |
Was Anyone Attacked?
Nobody has reported misuse, and for browsers there is a good reason to think there was none. Chrome, like other clients that enforce Certificate Transparency (CT), accepts a certificate only if it has been recorded in the public CT logs. Google: no TLS certificates chaining to these six intermediates were logged before its block, so "no fraudulent certificates issued by these CAs could have been used against CT-enforcing clients".
Outside the browser the picture is different. Mozilla's incident page: "many clients outside of the browser ecosystem do not enforce Certificate Transparency. We do not know if attackers exploited the flawed key generation to target these clients." That window opened in spring 2025 and closes for each client only when it is updated. eMudhra says it is not aware of any evidence that the keys were recovered before the disclosure, and Cloudflare says its review found no active exploitation on its services.
Now that the keys are public, forging a certificate takes no skill. Mozilla's incident page lists test servers that present certificates issued with the compromised keys, so that anyone can check their clients against them (the commands are below).
What a forged certificate allows: someone who can intercept the traffic between your software and a service it calls (on a shared network, through a compromised router or a poisoned DNS answer) can impersonate that service, read what your software sends, API keys and personal data included, and change what it receives. Software signed with one of the two code-signing keys would have passed the Windows signature check until Microsoft's block reached the machine.
Who Is Affected
There are three separate problems, and one company can have all of them.
1. Software that still trusts the compromised keys
Almost every company has this one: any program making HTTPS calls whose trust store still contains emSign ECC Root CA - G3 or AAA Certificate Services, and which has not received a block for the eight keys. Typical places:
- Linux servers and container images built before their distribution's update, which for most came between 20 and 24 October. An unpatched Debian 12, for example, trusts both chains.
- Language runtimes that ship their own list: Node.js, Python's certifi, Java's
cacerts, PHP's Composer bundle, Rust's webpki-roots. - Macs that have not installed the 9 November macOS updates. Apple says the built-in curl, git, python3, perl, ruby and openssl check certificates against
/etc/ssl/cert.pemwithout checking Apple's block or revocation. Since macOS 27.2.1, 26.7.3 and 15.8.3, those tools reject the compromised certificates, and the file no longer containsemSign ECC Root CA - G3,AAA Certificate Servicesor three other eMudhra roots,emSign Root CA - G1among them. - Windows machines that do not receive Microsoft's automatic updates to its disallowed certificate list, usually because Group Policy switches them off or the network is offline.
- Appliances, devices and third-party software that bundle their own CA file, such as load balancers, VPN gateways, SDKs and embedded controllers.
Each runtime has to be checked on its own. In a test we ran on 22 October, an unpatched Debian 12 machine with Node.js 24.18 accepted the chain that ends at eMudhra's root and rejected the one that ends at Sectigo's, while Python on the same machine accepted both.
2. Companies whose own certificates chain to the removed root
Mozilla, the Linux distributions and OpenJDK have removed emSign ECC Root CA - G3. Apple has blocked it since 23 October, the root lists of Windows and Edge dropped it on 26 and 27 October, and Chrome removed it in version 43 of its root store, scheduled to reach all devices on the morning of 27 October (US Pacific). Valid certificates under that root stop working in each client as it updates. Google's instruction: website operators with such certificates "must immediately acquire and install a new TLS certificate from another PKI hierarchy included in the Chrome Root Store".
Chrome counted about 30,800 unexpired certificates under the root in CT on 21 October. 30,791 of them came from two intermediates of InCommon, the certificate service Internet2 runs for US universities and research institutions, which issues ECC certificates through CERTInext, eMudhra's platform. 18 came from eMudhra's own intermediates. eMudhra says about 30,000 certificates need migrating, about 97% of them issued through ACME. In its answers to Chrome of 28 October it counted 25,058 still valid of the 29,349 valid when issuance stopped, and put the share actually replaced at about 70%, since many subscribers replaced a certificate without revoking the old one. It did not force replacement. Subscribers who used ACME, almost all in the US, moved to its emSign Root CA - G1 hierarchy at first, and the others got Sectigo certificates. On 12 November InCommon said it had moved issuance from eMudhra to SSL.com: DV certificates since 9 November, OV certificates once their validation is done.
If your site is behind Cloudflare in Full (strict) mode and the origin presents a certificate under this root, Cloudflare says the distrust "would cause breakages", and asks customers to "Review your custom origin certificates and server configurations".
RSA certificates from eMudhra's other root, emSign Root CA - G1, still work in browsers, but that root now has an end date. On 4 November Google announced that Chrome will no longer trust TLS certificates under it whose earliest Signed Certificate Timestamp (SCT), the proof that a certificate was logged in CT, is dated after 16 November 2026, 23:59:59 UTC; certificates logged by then are unaffected. Mozilla set the same cut-off in its trust store on 5 November (bug 2089115). Microsoft goes further: on 10 November it announced that it will remove G1 and eMudhra's nine other remaining roots from the Windows and Edge root lists on 16 November, and "Once these updates reach a device, the affected root certificates will no longer be trusted by default", whenever a certificate under them was issued. eMudhra says Apple will remove the root on 30 November, which would end trust in every certificate under it on Apple devices; Apple has not published that date itself, though its macOS updates of 9 November already remove G1 from the file the command-line tools use. Java does not wait either: OpenJDK's fix of 24 October removes emSign Root CA - G1 from Java's own trust store (cacerts), so an updated Java runtime that uses that store rejects those certificates (Temurin, which builds its store from Mozilla's list, keeps the root). On 22 October eMudhra announced that it will leave the TLS certificate business, and it offers its customers certificates issued by Sectigo instead. Its full report says all issuance from G1 will stop before 16 November. If you depend on eMudhra or CERTInext certificates of any kind, move to another certificate authority now: from 16 November, certificates under G1 stop working on Windows and in Edge as the update reaches each device.
3. GoDaddy customers
GoDaddy's case has nothing to do with eMudhra, but it hit the same week and many more sites. On 15 October GoDaddy reported to Mozilla that certificates from its GoDaddy and Starfield hierarchies carried certificate policy information (policy OIDs and CPS links) that did not match the profiles in its own CP/CPS, the document in which a certificate authority states what it issues. That is not a security flaw in the certificates. Under the Baseline Requirements, the rules every public certificate authority follows, it is still a reason to revoke them within five days.
GoDaddy first counted 404,607 affected certificates, 396,055 of them domain validated, and on 22 October corrected the total to 403,643. It revoked 170,566 by the deadline of 18 October, 23:45 UTC, and 227,422 by the next morning. On 22 October it reported the work finished: 378,067 revoked, 25,576 expired before they could be, none still valid. GoDaddy reissues the certificates, but its customer notice leaves installing them to the customer: "visitors to your website may see security warnings, and your site may become inaccessible until a new certificate is installed."
What to Check Today
1. What do your own servers send? For each public hostname, list the issuers in the chain it presents:
h=www.example.com
openssl s_client -connect "$h:443" -servername "$h" -showcerts </dev/null 2>/dev/null | grep -E '^ +i:'
A line naming eMudhra, emSign, one of the EM ... G3B or G3C authorities, or InCommon Intermediate CA - DVG3C, OVG3C or EVG3C means the certificate has to be replaced now: Chrome has removed emSign ECC Root CA - G3, Chrome and Firefox will not trust certificates first logged under emSign Root CA - G1 after 16 November, and Windows and Edge drop that root altogether on the same day. AAA Certificate Services on its own is normal, since many ordinary Sectigo chains still include it. For mail and directory services add -starttls smtp (or imap, ldap) and the right port. Include internal hosts, load balancers and origins behind a CDN.
2. What has been issued for your domains? CT lists every certificate a public certificate authority has issued for your names, including ones nobody in the company remembers ordering:
curl -s 'https://crt.sh/?q=%25.example.com&output=json&exclude=expired' | jq -r '.[].issuer_name' | sort | uniq -c
The output also shows which certificate authorities your company actually uses, which is the start of the inventory described further down.
3. Do your clients trust the compromised keys? Mozilla's test servers present certificates issued under the compromised keys, so a certificate error is the good result. Test both families: the first host chains to eMudhra's root, the second to Sectigo's, and a trust store can fail one and pass the other.
curl -sS -o /dev/null https://dv-g3b-emudhra.we-have-a-problem.com/ && echo "TRUSTED: update this system"
curl -sS -o /dev/null https://dv-g3b-aaa-emudhra.we-have-a-problem.com/ && echo "TRUSTED: update this system"
curl only answers for curl. Run the same request through each application's own TLS stack, in the environment where the application runs:
node -e "require('https').get('https://dv-g3b-emudhra.we-have-a-problem.com/', () => console.log('TRUSTED')).on('error', e => console.log('rejected:', e.code))"
python3 -c "import urllib.request; urllib.request.urlopen('https://dv-g3b-emudhra.we-have-a-problem.com/'); print('TRUSTED')"
python3 -c "import requests; requests.get('https://dv-g3b-emudhra.we-have-a-problem.com/'); print('TRUSTED')"
The two Python lines are not redundant: urllib uses the operating system's store, requests uses certifi. Repeat each line with the dv-g3b-aaa-emudhra host. For Java, make any HTTPS call to the test servers from the JVM in question, and list whether its store contains any of the four roots that OpenJDK's fix removes (Java 9 or later; on Java 8 use -keystore $JAVA_HOME/jre/lib/security/cacerts):
keytool -list -v -cacerts -storepass changeit | grep -E 'Owner: .*(emSign ECC Root CA - G3|emSign Root CA - G[12]|AAA Certificate Services)'
A root in a store does not prove exposure on its own, because Go, Node.js, Deno and Rust's rustls-webpki now block the eight keys whatever the store says. The test servers settle it.
4. Windows fleets. Microsoft confirmed on 26 October that the disallowed certificate list Windows downloads automatically has blocked all eight keys since 20 October, and that Edge and WebView2 received the block through a CRLSet update the same day. The exceptions are machines where Group Policy switches off automatic root certificate updates and networks with no route to Windows Update; those need the list delivered another way. Microsoft's check is certutil -f -verifyCTL DisallowedWU | findstr SequenceNumber, which should show 01dd5f6ca45c9d86 or a later number. A revision of the statement on 27 October adds that Microsoft removed six eMudhra G3 roots, emSign ECC Root CA - G3 among them, and AAA Certificate Services from the root list that Windows downloads on 26 October, and from Edge's on 27 October; once that reaches a device, those certificates are no longer trusted by default. On 10 November Microsoft announced the same for the remaining ten eMudhra roots, emSign Root CA - G1 among them, from 16 November. Mozilla's incident page offers a script for single machines and says it is "not recommended for enterprise-managed devices".
5. GoDaddy certificates. Check which certificate each server actually presents:
echo | openssl s_client -connect www.example.com:443 -servername www.example.com 2>/dev/null | openssl x509 -noout -issuer -serial -startdate
If the issuer is GoDaddy or Starfield, compare the serial number with the certificates in your GoDaddy account. Where GoDaddy shows "Install your new certificate", the server still has the revoked one.
One rule throughout: do not make an error go away by switching off certificate verification. Cloudflare's words: "It's important to not resolve these errors by disabling verification." A client that does not verify trusts every forged certificate, not only these.
What to Update This Week
There are two kinds of fix. Removing the two roots closes every path for a client that uses that store, but it also breaks the valid certificates under them. Blocking the eight keys by their public key hash, as Go, Node.js, Deno, Electron, rustls-webpki, Composer and Windows do, closes the hole and breaks nothing else. Where your platform offers the second, prefer it.
| Platform | Fixed in | Note |
|---|---|---|
| Debian 13 | ca-certificates 20250419+deb13u1 (DSA-6572-1) | |
| Debian 12 | ca-certificates 20250419~deb12u2 (DLA-4845-1) | |
| Ubuntu 22.04 to 26.10 | USN-8944-1; USN-9001-1, 5 November; USN-9036-1, 10 November | USN-9001-1 also removes the roots from Java's keystore; USN-9036-1 makes that work on systems set to certain languages, where it failed |
| Ubuntu 14.04 to 20.04 | USN-8944-2; for 18.04 and 20.04 also USN-9001-2, 9 November | Ubuntu Pro (ESM) only. USN-9001-2 also removes the roots from Java's keystore |
| Alpine 3.20 to 3.24 | ca-certificates 20261020 | Official images rebuilt on 21 October |
| Fedora 43 to 46 | ca-certificates updates (Red Hat bug 2556376) | |
| RHEL 7 (ELS) to 10 | ca-certificates in RHSA-2026:88455 (RHEL 10), RHSA-2026:88459 (RHEL 9), RHSA-2026:88550 (RHEL 8), RHSA-2026:88779 (RHEL 7 ELS), 22 and 23 October | Rebuild UBI-based images on a current base image |
| RHEL 6 (ELS), OpenShift (RHEL CoreOS) | RHSA-2026:90985 (RHEL 6 ELS); OpenShift 4 releases of 27 October, for 4.13 and 4.17 of 11 November (RHSB-2026-012) | Red Hat still lists OpenShift 5 as affected |
| Amazon Linux 2, 2023 and 2027 Preview | ca-certificates in ALAS2-2026-4037 and ALAS2023-2026-3306 (24 October) and ALAS2027-2026-262 (2 November) | Removes the eMudhra roots |
| FreeBSD | FreeBSD-SA-26:73.caroot | |
| Node.js | 22.23.4, 24.22.1, 26.12.1 | Also blocks the eight keys |
| Go | 1.26.10, 1.27.3 | Blocks the eight keys whatever the root store |
| Python | certifi 2026.10.19; pip 26.2.2 | The ssl module without certifi follows the operating system |
| Java | Releases that name JDK-8395690 (CVE-2026-114820, 24 October), such as Corretto 8.512.09.1, 11.0.33.0.10, 17.0.21.0.11, 21.0.13.0.11 and 25.0.5.0.10 | Removes four roots from cacerts, emSign Root CA - G1 among them. OpenJDK says to act "even if you have updated your runtime in the past few days"; without an update, delete the four roots as its eMudhra page shows. Temurin builds its store from Mozilla's list: Adoptium names 17.0.21+9, 21.0.13+9 and 25.0.5+8 as fixed, 8u512 and 11.0.33 followed on 4 November, and all of them keep emSign Root CA - G1. OpenJDK from the distribution follows the operating system's store, except on Debian, and on Ubuntu without USN-9036-1 (22.04 and later) or USN-9001-2 (18.04 and 20.04): rebuild the keystore with rm -f /etc/ssl/certs/java/cacerts and update-ca-certificates --fresh |
| PHP (Composer) | composer/ca-bundle 1.6.0 or 1.4.16 | |
| Rust | webpki-roots 1.0.10, rustls-webpki 0.103.16 | |
| Deno, Electron, Bun | Deno 2.9.8; Electron 42.11.16, 43.8.1, 44.9.1; Bun 1.4.4 | |
| Windows | Disallowed certificate list of 20 October | Protects only machines that download it. The root list of 26 October removes six eMudhra G3 roots, emSign ECC Root CA - G3 among them, and AAA Certificate Services; the remaining ten eMudhra roots, emSign Root CA - G1 among them, leave the Windows and Edge root lists on 16 November |
| macOS | Automatic block since 19 October; emSign ECC Root CA - G3 blocked since 23 October; macOS 27.2.1, 26.7.3 and 15.8.3 (9 November) for command-line tools | Those versions also remove AAA Certificate Services and four eMudhra roots, emSign Root CA - G1 among them, from /etc/ssl/cert.pem. For a Mac that cannot update, Apple's support note gives a workaround |
| Cloudflare One Client, WARP, cloudflared | Updates of 20 October; cloudflared 2026.10.2 | |
| AWS CLI | botocore 1.43.117; AWS CLI v2 2.38.4 | botocore uses certifi when it is installed; AWS CLI v2 drops AAA Certificate Services from its own bundle |
The table is as of 13 November. Check again before you plan around it.
Four things catch teams out:
- Container images keep the old list. An image built before 20 or 21 October trusts the keys until it is rebuilt on an updated base image, and pulling the same tag helps only once the base image has been rebuilt. Rebuild, redeploy, and ask the vendors whose images you run.
- Processes keep what they loaded. A program reads its trust store when it starts, so restart long-running services after updating. Node.js adds that TLS sessions resumed from before the upgrade are not checked again.
- Applications carry their own CA files. SDKs and applications often ship a
cacert.pemorca-bundle.crtof their own. Search repositories and images for them, for example withfind / -xdev \( -name 'cacert*.pem' -o -name 'ca-bundle*.crt' \) 2>/dev/null, and expect more than one: pip, for instance, carries its own copy of certifi. - Devices wait for their vendor. Ask each vendor in writing which trust store the product uses and when the update ships. Until it does, the device trusts forged certificates for every service it calls, so limit what it can reach.
If you remove AAA Certificate Services by hand, or an update removes it (Debian, Ubuntu ESM, Fedora, OpenJDK, Temurin, AWS CLI v2, the root lists of Windows and Edge and the file macOS command-line tools use have all dropped it), test connections to services with Sectigo certificates afterwards: some older servers still send a chain that ends at that root.
What to Put in Place Before the Next One
A certificate authority leaving keys this weak is rare. Mass revocations are not. These are the large ones of 2026 recorded in Mozilla's bug tracker:
| Certificate authority | Month (2026) | Certificates affected | Reason |
|---|---|---|---|
| IdenTrust | June | 42,858 | Policy OIDs that its CP/CPS no longer allowed |
| HARICA | July | 63,525 | Client authentication usage after the cut-off in its own CP/CPS |
| HARICA | July | 235,301 | No OCSP address, which its own CP/CPS required |
| Actalis | August | about 260,000 | Client authentication usage that its CPS excluded |
| GoDaddy | October | 403,643 | Policy information that did not match its CP/CPS |
| TWCA | November | 52,954 | Serial numbers with less randomness than the rules require, in its intermediate and subscriber certificates |
None of the first five was a security failure: in each, a certificate disagreed with the authority's own document. TWCA broke a rule on the randomness of serial numbers, not a key: its intermediates carried about 24 random bits and its TLS and S/MIME certificates 63, where the rules require at least 64. It revoked them all by 5 November, and its full report of 12 November says it "found no indication that the insufficient entropy was exploited". The Baseline Requirements (section 4.9.1.1) still require revocation within five days. When a key is compromised, or domain validation cannot be relied on, the deadline is 24 hours. Nobody can extend it for you. In DigiCert's 2024 revocation of 83,267 certificates, the Chrome Root Program wrote that it "does not have the authority to grant exceptions"; one customer obtained a court order to stop the revocation of its certificates, and the order was vacated four days later.
How the replacements went says a lot. In July HARICA used ACME Renewal Information (ARI) to tell its customers' ACME clients to renew early, and reissued other certificates itself; in the larger of its two revocations, 87% of the affected names had been replaced when the deadline came. GoDaddy, whose customers download and install each replacement themselves, had revoked 42% of its affected certificates by its deadline.
The renewal cycle is getting shorter in any case:
- Under CA/Browser Forum ballot SC-081, public TLS certificates have been limited to 200 days since 15 March 2026. The limit drops to 100 days on 15 March 2027 and to 47 days on 15 March 2029, when domain validation can be reused for only 10 days.
- Let's Encrypt moves its default to 64-day certificates on 10 February 2027, and to 45 days a year later.
- Public TLS certificates are losing client authentication. Let's Encrypt stopped issuing it on 8 July 2026, and the Chrome Root Program requires certificates issued from 15 March 2027 to be for server authentication only. If you use public certificates as client certificates for mutual TLS, which is common in B2B API integrations, those connections need a private certificate authority.
At 100 days, and then at 47, installing certificates by hand stops being a realistic option. A company that automates for that schedule is also ready for the next mass revocation. What that takes:
- Know what you have, in both directions. One list of every certificate you serve, with hostname, issuer, expiry, renewal method and owner. A second list of every trust store your software relies on: the operating system package, each language bundle, each CA file inside an application or device. The CT search above starts the first list; a software bill of materials (SBOM) or dependency inventory starts the second.
- Automate issuance and installation. ACME wherever it can run, with ARI (RFC 9773) so that a certificate authority can tell your clients to renew early. Where a device cannot run ACME, use the vendor's automation, or terminate TLS in front of it on something that can.
- Keep a second certificate authority ready. A working ACME account with a second authority, tested, and CAA records in DNS that allow both. When one authority is distrusted or revokes in bulk, switching becomes a configuration change.
- Monitor CT for your domains. Get an alert for every new certificate for your names. You learn about unexpected issuance the day it happens.
- Treat trust stores as dependencies. Update them through normal patching, and rebuild images on a schedule rather than only when code changes.
- Write the runbook against the two clocks. 24 hours and five days: who decides, who has access to each system, which systems need manual steps, how to reach each vendor. Rehearse it once a year by swapping the certificate on a service that can afford a mistake. Public certificate authorities have had to keep and test a mass revocation plan of their own since 2025; their customers need the other half.
If You Make Products That Others Run
If your product contains a trust store (a device, an installed application, an SDK), your customers are protected only once your update reaches them. Ship it, and tell customers which versions are fixed. Mozilla's advice to trust store operators: block the eight intermediates if your software can, otherwise "consider whether removing trust in the roots which signed these intermediates is suitable".
The Cyber Resilience Act applies here in two ways. Its reporting obligation, in force since 11 September 2026, covers actively exploited vulnerabilities, and no exploitation has been reported in this case. Its essential requirements, which apply from 11 December 2027, include that vulnerabilities can be addressed through security updates; a product whose trust list cannot be updated would not meet them. Customers covered by NIS2 will ask their suppliers how quickly they handled this, and it helps to have the answer written down.
What Is Still Open
- eMudhra's remaining root. Chrome and Firefox will not trust certificates first logged under
emSign Root CA - G1after 16 November, and Windows and Edge remove the root that day. Google warns that "the date and approach described above may shift if new information comes to light". eMudhra says Apple will remove the root on 30 November. OpenJDK has already removed it from Java's own trust store. - eMudhra's exit. eMudhra will leave the TLS certificate business and offers certificates issued by Sectigo. On 5 November it said it had moved every emSign TLS customer outside the US to other providers. Its full report says all issuance from G1 will stop before 16 November, and InCommon, by far its largest TLS customer, has moved to SSL.com.
- Pending fixes. By 13 November Red Hat still listed the RHEL CoreOS of OpenShift 5 as affected. Every other platform in the table had a fix.
- The review. eMudhra's full report explains the cause, and an independent review of its key handling is due by 4 December 2026. Participants in Mozilla's incident bug have asked it to publish the script and the ceremony logs.
- Sectigo. In a separate case, open since 5 September, Sectigo has declined to revoke certificates whose extended key usage did not match a sentence in its CPS, arguing the sentence was ambiguous. Its latest count, for 12 November, was about 38.1 million of them still active. Some participants in the public discussion have called for distrust; no root program has announced a decision. Companies with Sectigo certificates should follow the case.
Where to Get Help
A certificate problem rarely announces itself as one. It shows up as a checkout that stops loading, an integration that stops syncing or a device that stops reporting, usually on a weekend. We help companies get through weeks like this one without an outage, and set things up so the next one is routine:
- An exposure check. We find where your systems make or accept TLS connections, which trust store each one uses and which certificates you serve, including containers, Java services, integrations with banks, payment and logistics providers, and devices.
- The fixes. Updates, rebuilt images, blocked keys or replaced certificates, made through your change process and confirmed against Mozilla's test servers.
- Automation for 2027. ACME with ARI across servers, load balancers and appliances, a second certificate authority, CAA records, CT monitoring and a runbook, ready for 100-day certificates from 15 March 2027.
- The systems nobody wants to touch. Old Java applications, appliances and embedded clients with their own trust stores are what our legacy system and code maintenance service is for.
If you are not sure whether you are affected, write to office@c9group.dev with the domains and platforms you run, or book a call. We will tell you what to check first.
Cybersecurity
New guides and rule changes by email
When a rule changes, a deadline moves or we publish a new guide, we send it by email.
Knowledge Hub
Related articles

Cybersecurity
The Cyber Resilience Act: Reporting Has Applied Since 11 September 2026
Reporting has applied since 11 September 2026, with a 24 hour clock. Scope, SBOM pipelines, vulnerability monitoring, incident runbooks and full application in December 2027.

Cybersecurity
NIS2 in Practice: Registration, 24 Hour Reporting and Supply Chain Reality
Who is in scope, registration duties, the 24 hour early warning, the ten required measures, and what customers now ask suppliers to prove.

Cybersecurity
PHP2Shell (CVE-2026-109618 and CVE-2026-109619): What Companies Running PHP Should Do Now
Two critical PHP flaws let an attacker run code on a web server with a single request. Who is affected, which versions have a fix, what to do if you cannot patch today, how to check for a break-in, and what you may have to report.