We earn commissions when you shop through affiliate links on this site, at no extra cost to you. Our guides and technical recommendations are written independently of these relationships: a commission never decides what we recommend, how we diagnose a problem or what we publish, and the content does not depend on it. Affiliate disclosure

Security · Published 2026-11-04

CVE-2026-114820: eMudhra Java Trust Store Fix

CVE-2026-114820 removes the eMudhra roots from Java trust stores. Update Ubuntu, regenerate the Java store and verify it with the vendor test hosts.

TL;DR

  • Eight eMudhra intermediate CA certificates have private keys that anyone can recover from the certificates themselves. An attacker who holds one of those keys can issue a certificate for any hostname, and a client that still trusts the chain accepts it.
  • Java got two fixes. CVE-2026-91403 (21 October) blocked the eight intermediates. That was not enough, so CVE-2026-114820 (24 October) removes the roots that signed them from the trust store. A runtime with only the first update is not done.
  • Fixing the operating system does not fix Java on Ubuntu. We tested it: after the ca-certificates update alone, Ubuntu's generated Java store still held the emSign ECC Root CA - G3 root, and a Java client still accepted both vendor test hosts while curl rejected them. After Canonical's two extra commands the root was gone and Java rejected them too.
  • Oracle's .0.2 builds are not the fix. Oracle's alert lists 17.0.21.0.2, 21.0.13.0.2 and the other .0.2 builds under "Affected Products and Versions". Oracle's fixed builds are in its patch availability document, which we did not read, so this post does not name them.
  • A standalone JDK, a container image or an application with its own trust store needs its own fix. Patching the host changes none of them.
  • RHEL, AlmaLinux, Rocky: the package update is enough for the distribution's OpenJDK. There the Java store is a link to the system trust store, so dnf upgrade ca-certificates fixes Java as well. We tested this on AlmaLinux 10.2 and in an almalinux:10 container.
  • Test with the vendor's hosts, not with a version string. Four test hostnames are published by the incident tracker; a certificate error means the client no longer trusts the compromised chains.

How this was checked

The facts below were read from the vendor and incident pages listed under Sources on 4 November 2026.

The commands marked tested were run on 4 November 2026 on a disposable Ubuntu 26.04.1 LTS test machine that still had the vulnerable ca-certificates 20260601 (OpenJDK 21.0.12.1 installed from Ubuntu's repository), and on an ubuntu:24.04 container. The test used the four hostnames from the incident tracker, which only serve the test chains, and a ten-line Java program of our own that opens an HTTPS connection. No exploit code was used.

The RHEL-family variants were run on 10 November 2026 on an AlmaLinux 10.2 test machine and in an almalinux:10 container, both already on the fixed ca-certificates, with Red Hat-family OpenJDK 21 (java-21-openjdk-headless). There was no vulnerable RHEL-family build to compare against, so those tests show the fixed state only.

Not run: the vendor's CheckCerts.java checker (we did not execute downloaded code), the Oracle, Temurin and Red Hat Build of OpenJDK binaries, Ubuntu 22.04 and 24.04 hosts (only a 24.04 container), the revocation-checking options, and anything on a live Kubernetes cluster. The last section lists what that leaves open.

Timeline

  • 30 April 2025: eMudhra says the eight intermediates were created in a key-generation ceremony with a flawed step.
  • 17 October 2026: a researcher reports weak keys. Mozilla received the report at 20:06 UTC and had blocked the certificates in Firefox through OneCRL at 22:59 UTC, in under three hours.
  • 18 October: eMudhra revokes all eight intermediates.
  • 20 October: Ubuntu publishes the ca-certificates update (USN-8944-1). Microsoft platforms block the certificates the same day.
  • 21 October: Ubuntu publishes the NSS update (USN-8947-1). OpenJDK and Oracle publish CVE-2026-91403, the intermediate block.
  • 22 October: Apple products block the certificates.
  • 24 October: OpenJDK and Oracle publish CVE-2026-114820, the root removal.
  • 30 October: Adoptium releases Temurin 17, 21, 25 and 27 with the fix. eMudhra appoints KPMG as an independent reviewer and says its root-cause analysis will follow.

What happened

A TLS chain runs from the website's certificate through an intermediate CA to a trusted root. The intermediate signs the leaf certificate, so whoever has the intermediate's private key can sign a certificate for any name.

eMudhra says a step in its key ceremony "produced a second, weak set of keys alongside the original strong keys", and that the weak keys are the ones that were published. The incident tracker says anyone can recover the private keys from the certificates. eMudhra states that none of the intermediates issued publicly trusted certificates and that its infrastructure and customer data were not compromised, and that it is not aware of evidence that the keys were exploited before disclosure. The root cause was not final at its 30 October update.

Certificate Transparency protects browsers, not every client. The intermediates have no Certificate Transparency entries for the relevant period, so clients that enforce Certificate Transparency would reject a forged certificate. The tracker warns that many non-browser clients, curl among them, do not enforce it, and says it does not know whether attackers went after those. We saw this ourselves: a Java client and curl both accepted the vendor's "no-CT" test host on the vulnerable machine.

The tracker also reported on 28 October that the developer tools Apple ships with macOS, including git, curl and python3, still trusted the certificates, and that macOS 27.2 undid user-applied fixes. This post is about Ubuntu, Java and containers, so check your macOS fleet separately.

The two Java CVEs

CVEPublishedWhat the Java change didConclusion
CVE-2026-9140321 OctoberBlocked the eight intermediate certificatesRed Hat says it "proved insufficient and easy to bypass"
CVE-2026-11482024 OctoberRemoves the roots that signed the intermediates from the trust storeApply this one

Both are scored CVSS 3.1 9.1 in the OpenJDK advisories and in Oracle's risk matrix. Red Hat scores CVE-2026-114820 at 8.1 and CVE-2026-91403 at 6.8; its pages show no severity label. The scores differ because the scorers' assumptions differ. The weakness is improper certificate validation: a Java TLS client can accept a fraudulent chain.

Why patching the host does not patch every Java runtime

A process can use any of these stores, and the fix has to reach the one it actually uses:

  • the system CA bundle (OpenSSL-based clients, curl)
  • NSS data (Firefox and other NSS programs)
  • a Java cacerts file, which may be a copy generated from the system bundle or one that ships inside the JDK
  • a custom store named by javax.net.ssl.trustStore
  • a bundle baked into a container image or a statically linked program

How each Java distribution gets its roots differs. RHEL's OpenJDK packages rely on the system ca-certificates package, while Red Hat Build of OpenJDK for Linux and Windows carries its own store and is updated separately. Temurin builds its store from Mozilla's CA data. Oracle JDK ships its own.

1. Update Ubuntu's system CA bundle and NSS

Canonical's fixed ca-certificates versions:

Ubuntu releaseFixed ca-certificates
22.04 LTS20260816ubuntu0.22.04.1
24.04 LTS20260816ubuntu0.24.04.1
26.04 LTS20260816ubuntu0.26.04.1
26.1020260816ubuntu0.26.10.1

Releases 14.04 to 20.04 need Ubuntu Pro or the legacy support add-on; see USN-8944-2 and USN-8947-1 for their versions. The NSS (libnss3) fixes are 2:3.98-0ubuntu0.22.04.5 on 22.04, 2:3.98-1ubuntu0.3 on 24.04, 2:3.120-1ubuntu2.2 on 26.04 and 2:3.127-1ubuntu0.1 on 26.10.

The ca-certificates update removes emSign ECC Root CA - G3. The NSS and ESM updates also remove the Comodo AAA root, which Ubuntu had already dropped from the 22.04 and newer bundles. (Canonical's blog post calls the first root "emSign Root TLS CA – G3"; the security notices and the Java tools use the name above.) Canonical warns that any certificate whose chain ends in either removed root will no longer validate.

Tested. This is the update and the check on the vulnerable machine:

# Check the versions before and after; 'libnss3*' matches only if the package is installed.
dpkg -l ca-certificates 'libnss3*' | grep '^ii'
sudo apt-get update
sudo apt-get install --only-upgrade ca-certificates 'libnss3*'

Before, ca-certificates was 20260601~26.04.1. After, it was 20260816ubuntu0.26.04.1 and libnss3 was 2:3.120-1ubuntu2.2, both the fixed versions. Java had pulled libnss3 in as a dependency; on a server without it, the glob simply matches nothing.

To inspect the system bundle:

# Prints a line for each match; no output means the root is gone (grep exits 1).
openssl crl2pkcs7 -nocrl -certfile /etc/ssl/certs/ca-certificates.crt |
  openssl pkcs7 -print_certs -noout |
  grep -i -E 'emSign ECC Root CA - G3|AAA Certificate Services'

Before the update this printed the emSign ECC Root CA - G3 subject and issuer lines; after it, nothing, and the bundle went from 121 to 120 certificates. It printed no AAA line either time, because Ubuntu 26.04's bundle no longer had that root.

Use the vendor's test hosts for the real answer:

# A certificate error means this client no longer trusts the compromised intermediates.
curl -sS -o /dev/null -w '%{http_code}\n' https://dv-g3b-emudhra.we-have-a-problem.com/
curl -sS -o /dev/null -w '%{http_code}\n' https://dv-g3b-noct-emudhra.we-have-a-problem.com/

On the vulnerable machine both returned 200. After the update both failed with curl: (60) SSL certificate OpenSSL verify result: unable to get local issuer certificate. The tracker also lists two "aaa" hosts for the Comodo chain; on the vulnerable machine both already failed, because Ubuntu 26.04’s bundle did not have that root.

RHEL, AlmaLinux, Rocky. The fix is in ca-certificates 2026.2.92_v9.0.319-100.1 and later; its changelog lists "Removing: AAA Certificate Services root CA" and "Removing: emSign ECC Root CA - G3". Check and update:

# Installed version, and whether its changelog names the removed roots
rpm -q ca-certificates
rpm -q --changelog ca-certificates | grep -i -E 'emSign ECC Root CA - G3|AAA Certificate Services'
sudo dnf upgrade ca-certificates
# No output means neither root is a trust anchor any more
trust list --filter=ca-anchors | grep -i -E 'label: (emSign ECC Root CA - G3|AAA Certificate Services)'
# Real output from the test machine (AlmaLinux 10.2, 10 November 2026), shortened
ca-certificates-2026.2.92_v9.0.319-100.3.el10_2.noarch
- Removing: AAA Certificate Services root CA
- Removing: emSign ECC Root CA - G3 from Microsoft Code Signing certificates

The last command printed nothing. The other emSign roots (G1, G2, C2, C3 and the code-signing CS G2) are still listed, as on Ubuntu. Both vendor test hosts failed in curl with (60) SSL certificate problem: unable to get local issuer certificate. The system bundle on RHEL is /etc/pki/tls/certs/ca-bundle.crt, if you prefer the openssl check above with that path.

2. Regenerate the Java store on Ubuntu and Debian

Canonical says Ubuntu's OpenJDK uses the system store, but "due to an unrelated issue" (Launchpad bug 2061817) the Java store needs manual action. The OpenJDK guidance says the same for Debian and its derivatives: updating ca-certificates and ca-certificates-java is not enough.

Tested. We installed openjdk-21-jre-headless before the update and looked at the store afterwards:

  • Right after the ca-certificates update, /etc/ssl/certs/java/cacerts still contained debian:emsign_ecc_root_ca_-_g3.pem, even though apt had run the ca-certificates-java trigger.
  • A Java program of ours still got HTTP 200 from both vendor test hosts at that point, while curl already refused them.

Canonical's fix, with a backup first:

# Back up the store: the next command removes any locally added certificates.
sudo cp -a /etc/ssl/certs/java/cacerts \
  /etc/ssl/certs/java/cacerts.backup.$(date +%Y%m%d%H%M%S)
sudo rm -f /etc/ssl/certs/java/cacerts
sudo update-ca-certificates --fresh

Canonical says certificates added directly to the Java store are removed by this and must be re-added through your normal process. We had none to lose, so that part is Canonical's statement, not our test. After the commands the emsign_ecc_root_ca_-_g3 alias was gone (the other eMudhra roots, emsign_root_ca_-_g1 among them, were still there), and the Java program now failed for both test hosts with SSLHandshakeException ... PKIX path building failed, while example.com still returned 200.

Restart every Java process after this. A running JVM keeps the trust managers it started with.

RHEL, AlmaLinux, Rocky: no regeneration step. The distribution's OpenJDK reads /etc/pki/ca-trust/extracted/java/cacerts, which update-ca-trust writes from the same sources as the system bundle (the package update runs it). Tested on AlmaLinux 10.2 with java-21-openjdk-headless 21.0.13:

# Which store this java reads, and whether the removed root is still in it
readlink -f "$(dirname "$(readlink -f "$(command -v java)")")/../lib/security/cacerts"
keytool -list -cacerts -storepass changeit | grep -i emsign
# Real output from the test machine
/etc/pki/ca-trust/extracted/java/cacerts
emsignrootca-g1, Nov 8, 2026, trustedCertEntry,

Only the G1 root is left, which is expected. Our Java program failed for both test hosts with PKIX path building failed and reached example.com. Red Hat documents that certificates added through /etc/pki/ca-trust/source/anchors/ are included when the store is rebuilt, so local additions made that way survive; we had none to check. Red Hat Build of OpenJDK installed from a tarball, and any other JDK with its own cacerts, is not covered by this: see the next sections.

3. Standalone Oracle JDK and GraalVM

Oracle's CVE-2026-114820 alert, published 24 October, describes Java updates "that remove the set of impacted root CA certificates". It also lists Oracle Database, Autonomous Health Framework, OPatch and other products that embed Java.

The versions Oracle lists in the alert are the affected ones: Java SE 8u511-b32, 11.0.33.0.2, 17.0.21.0.2, 21.0.13.0.2, 25.0.5.0.2 and 27.0.1.0.2, and the same .0.2 builds of GraalVM for JDK 17 and 21. The OpenJDK advisory agrees: it lists 27.0.1.0.2, 25.0.5.0.2, 21.0.13.0.2, 17.0.21.0.2, 11.0.32.1 and 8u504 and earlier as affected. These are the builds that carry the first, intermediate-only fix. Oracle's fixed builds are in its patch availability document; require the release notes of the build you install to say that it fixes CVE-2026-114820 (OpenJDK issue JDK-8395690).

Check which runtime a service really starts, not only the one on the PATH:

# A host can have a patched /usr/bin/java and an older JDK bundled with an application.
java -version
readlink -f "$(which java)"

4. Manual workaround while patching is delayed

OpenJDK's guidance, referenced from Oracle's alert, is to remove four roots from the runtime's cacerts. Back it up first.

# JDK 11 and later (JDK 8: add -keystore $JDK_HOME/jre/lib/security/cacerts). Ignore "does not exist" errors.
keytool -delete -cacerts -alias "emsignrootcag1 [jdk]"
keytool -delete -cacerts -alias "emsignrootcag2 [jdk]"
keytool -delete -cacerts -alias "emsigneccrootcag3 [jdk]"
keytool -delete -cacerts -alias "comodoaaaca [jdk]"

Restart the running instances afterwards, and repeat it for every standalone installation.

Two cautions we found. First, these aliases are the ones in the JDK's own store. Ubuntu's generated store names them differently (debian:emsign_ecc_root_ca_-_g3.pem), and tested on a copy of the store, the emsigneccrootcag3 [jdk] command failed with Alias <emsigneccrootcag3 [jdk]> does not exist. On Ubuntu, regenerate the store as in section 2 instead. Second, vendors do not agree on the list: Temurin keeps emSign Root CA - G1 because Mozilla still trusts it, and Adoptium says the upstream checker wrongly flags it as a bad root. Follow the guidance of the vendor of the runtime you actually run.

OpenJDK's other option for TLS applications is to turn on revocation checking:

# Options for the java command line; behavior changes when revocation data cannot be fetched.
-Dcom.sun.net.ssl.checkRevocation=true
-Dcom.sun.security.enableCRLDP=true

For signed JARs on JDK 17 and later it documents jarsigner -verify -revcheck -strict <jar>. We did not run either. Test them with your application's network conditions, because a blocked revocation lookup can break connections.

5. Temurin and other OpenJDK builds

Adoptium says the Temurin releases of 30 October contain the fix: JDK 17.0.21+9, 21.0.13+9, 25.0.5+8 and 27.0.1.0.2+1. It builds the store from NSS 3.131.1 data, which no longer has emSign ECC Root CA - G3 and AAA Certificate Services. As of that release note, JDK 8 and 11 were still waiting for upstream tags.

Version numbers do not line up between vendors. Oracle and OpenJDK list 17.0.21.0.2 as affected while Adoptium lists 17.0.21+9 as fixed, and Adoptium says its numbers do not match later upstream tags. Read each vendor's release note and look for CVE-2026-114820, not a familiar version number.

6. Containers and Kubernetes

The host's trust store is irrelevant to a container that carries its own. Start with an inventory of the images that are running.

The pipeline below includes init containers, which a loop over .spec.containers alone would miss.

# Lists namespace, Pod, container and image, including init containers.
kubectl get pods -A -o json |
jq -r '.items[] as $pod
  | (($pod.spec.initContainers // []) + $pod.spec.containers)[]
  | [$pod.metadata.namespace, $pod.metadata.name, .name, .image] | @tsv'

Tested: the jq part, against a sample Pod list with an init container (the busybox init container appeared). We did not run it against a live cluster.

An Ubuntu-based Java image should install the current packages and regenerate the store:

# Ubuntu-packaged OpenJDK only. It does not fix a standalone JDK copied into the image.
FROM ubuntu:24.04

RUN apt-get update \
 && apt-get install -y --no-install-recommends \
      ca-certificates \
      libnss3 \
      ca-certificates-java \
      openjdk-21-jre-headless \
 && rm -f /etc/ssl/certs/java/cacerts \
 && update-ca-certificates --fresh \
 && rm -rf /var/lib/apt/lists/*

WORKDIR /app
COPY app.jar /app/app.jar

CMD ["java", "-jar", "/app/app.jar"]

Tested: this built, with an empty app.jar. Inside the image ca-certificates was 20260816ubuntu0.24.04.1 and libnss3 was 2:3.98-1ubuntu0.3, the fixed versions for 24.04. The Java store held no emsign_ecc_root_ca_-_g3 alias, and our Java program was rejected by both test hosts and reached example.com.

For a RHEL-family image, updating the package is enough, because the distribution's OpenJDK reads the system store:

# RHEL-family OpenJDK reads the system store, so updating ca-certificates is enough.
FROM almalinux:10

RUN dnf -y upgrade ca-certificates  && dnf -y install java-21-openjdk-headless  && dnf clean all

Tested: built and run with Docker on the Ubuntu 26.04 machine. Inside the image ca-certificates was 2026.2.92_v9.0.319-100.3.el10_2, the JDK's cacerts resolved to /etc/pki/ca-trust/extracted/java/cacerts, and our Java program was rejected by both test hosts and reached example.com.

For an Oracle JDK or any self-contained runtime, replace the runtime with the fixed vendor build instead, and keep your approved private CA additions in the build. Rebuild, publish a new image digest, and roll it out. A Pod is not patched because its node was.

7. Validate the store the process uses

A package version is not proof. Check the trust store the process really reads:

  1. Confirm the JDK vendor and the exact build.
  2. Confirm that its release note says it fixes CVE-2026-114820.
  3. Inspect the store the process is configured with, not an arbitrary JDK on the host.
  4. Run a controlled TLS test from the same runtime and network path. A connection through a TLS-inspecting proxy tests the proxy's store, not the client's.
  5. Confirm the service restarted.
  6. Test your ordinary HTTPS endpoints too, because removing a root can break legitimate chains that end in it.

OpenJDK recommends its CheckCerts.java program for step 4. We did not run it; the Java connection test above does the same job for the two chains we cared about.

Common mistakes

  • Stopping at CVE-2026-91403. The intermediate block was the first fix, and it was judged bypassable.
  • Patching the host, not the image. The container has its own store.
  • Updating ca-certificates and leaving Ubuntu's Java store stale. We reproduced it.
  • Assuming all JDKs are alike. Oracle JDK, Temurin, RHEL's OpenJDK and Red Hat Build of OpenJDK get their roots in different ways.
  • Using the [jdk] aliases on a generated store. They do not exist there.
  • Trusting a browser test. Browsers get revocation updates and enforce Certificate Transparency in ways Java and curl do not.
  • Deleting roots without a compatibility test. Chains that end in a removed root stop working.

If you suspect interception

A client with a vulnerable store would accept a forged chain, and that leaves few application-level traces. Keep TLS client, proxy and application logs for the exposure window. Find out which runtime and which store the process used at the time. Look for unexpected certificate chains and unusual destinations, and correlate them with DNS, routing and endpoint events. If compromise looks credible, contain the system, keep the evidence and rebuild from a trusted image. Rotate secrets that may have crossed an intercepted session, based on the evidence. This article is not incident-response advice for your environment.

Checklist

  • [ ] Find the hosts, images and runtimes that carry a trust store, and which store each Java process uses.
  • [ ] Apply the fixed Ubuntu ca-certificates and NSS packages; confirm with dpkg -l ca-certificates 'libnss3*'.
  • [ ] On RHEL, AlmaLinux and Rocky, apply ca-certificates 2026.2.92_v9.0.319-100.1 or later with dnf upgrade ca-certificates.
  • [ ] On Ubuntu and Debian hosts with Java, back up and regenerate the Java store. (Not needed for the RHEL-family OpenJDK packages.)
  • [ ] Update Oracle JDK and GraalVM to a build whose notes name CVE-2026-114820, not only CVE-2026-91403.
  • [ ] Update other JDK vendors from their own release notes; track JDK 8 and 11 where the vendor is late.
  • [ ] Rebuild images and self-contained runtimes and deploy the new digests; restart every Java process.
  • [ ] Run the vendor test hosts from the same runtime and network path.
  • [ ] Test your legitimate HTTPS endpoints for chains that ended in a removed root.

What we have not verified

  • Oracle's fixed build numbers. They are in a patch availability document we did not read.
  • The vendor checker. CheckCerts.java was not run.
  • Oracle JDK, Temurin and Red Hat Build of OpenJDK binaries. We tested the distribution OpenJDK 21 packages of Ubuntu and AlmaLinux only.
  • Ubuntu 22.04 and 24.04 hosts. We tested a 26.04 machine and a 24.04 container.
  • The RHEL path on a vulnerable build. We ran it on AlmaLinux 10.2, which was already fixed, so we saw the fixed state only. RHEL 8 and 9 were not tested.
  • Revocation checking and jarsigner -revcheck.
  • A live cluster.
  • The final root cause. eMudhra had not published it at its 30 October update.

If something differs on your system, tell us at info@opsbreakfix.com and we will fix the page.

Related guides

Sources

More from the blog

This article is educational information, not incident-response or legal advice. Verify package status with your operating system, hosting, or runtime vendor.