CVE monitoring and patch operations instead of finding out by chance - WZ-IT records where Java runs in your environment and where the JDK comes from, sets the workaround and applies updates. Book an initial consultation · Server audit · Keycloak operations
On 17 November 2026 Oracle and the OpenJDK project released updates for CVE-2026-114938, called Java4Shell in reports. A flaw in the optimized Base64 decoder of OpenJDK 17, 21, 25 and 27 allows memory beyond the target buffer to be overwritten on x86-64 processors with AVX-512 VBMI. The discoverers at Calif report code execution without authentication for Keycloak and Jenkins and after authentication for Elasticsearch, Apache Kafka, Apache Solr and Gerrit.
Because the flaw sits in the runtime, it is not the version of the application that matters but the JDK underneath, and for each application that JDK comes from a different place. This article classifies common open-source applications, names the origin of their JDK, separates what is documented from what is assumed and describes workaround, verification and update.
Table of contents
- In short
- Status of the information
- When the flaw applies
- Which open-source applications are affected
- Where your application's JDK comes from
- Patch status of the JDK vendors
- Workaround: switching off the Base64 intrinsic
- What to do now
- Is the flaw being exploited?
- Timeline
- Common misconceptions
- Status and methodology
- Our approach at WZ-IT
- Further guides
In short
| Question | Answer |
|---|---|
| Identifier | CVE-2026-114938, called Java4Shell in reports; component hotspot/compiler |
| Cause | Out-of-bounds write in the x86-64 Base64 intrinsic with AVX-512 VBMI: with malformed padding the decoder writes up to 62 bytes beyond the target buffer (Calif, Datadog) |
| Severity | CVSS 3.1: 8.1 in the OpenJDK advisory; CVSS 4.0: 9.2 according to Calif; Red Hat provisionally 9.8 |
| Affected | OpenJDK from 17.0.6 and 21, 25 and 27 up to and including 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3 and 27.0.1.0.3; plus Oracle Java 8 Performance Pack 8u511b33-perf |
| Not affected | Java 8 and 11 by default, ARM64, x86-64 without AVX-512 VBMI |
| Patched | Oracle 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1 and 8u513-perf; details per vendor further below |
| Workaround | -XX:-UseBASE64Intrinsics, then restart; alternatively -XX:UseAVX=2 on x86-64 only |
| Exploitation | No confirmed exploitation, not in the CISA KEV catalog; application-specific exploit chains are not published |
Status of the information
As of 18 November 2026, 2 pm CET. The situation keeps changing: vendors publish builds, projects respond with releases, and the discoverers have announced that they will release exploit chains later. The main sources are the OpenJDK advisory of 17 November, the Calif publication, the Red Hat bulletin RHSB-2026-018, the notice from the Keycloak team and the Adoptium project's post on Temurin. Which statements rest on vendor sources and which on third-party reports is explained in the section Status and methodology.
When the flaw applies
Three conditions have to be met at the same time (Keycloak):
- Unpatched OpenJDK from 17.0.6, including 21, 25 and 27. Java 8 and 11 are not affected by default, according to Adoptium and the OpenJDK advisory.
- An x86-64 processor with AVX-512 VBMI. According to Calif and Azul these are Intel Xeon (Ice Lake, Sapphire Rapids, Emerald Rapids, Granite Rapids) and AMD EPYC and Ryzen from Zen 4 onwards. ARM64 is not affected.
- Active Base64 intrinsics, that is, no limit of the JVM to AVX2 and no disabled intrinsic.
In addition there has to be an entry point: data from an attacker has to reach the Base64 decoder of the JVM. According to Datadog, in the tested build the Basic and the URL decoder are affected, the MIME decoder is not. Servers often decode Base64 before they check the authentication, for example for HTTP Basic logins, SAML artifacts and CSRF tokens (The Hacker News). A Java client that opens a TLS connection to an attacker's server can also be affected (Calif).
Calif distinguishes two effects: reading memory, such as keys and tokens, and executing code on the server. A failed attempt can crash the JVM. There is no universal exploit; each application needs its own attack chain.
Check the CPU. On Linux this command shows whether the host has the required features. All four flags must appear (Calif):
grep -m1 '^flags' /proc/cpuinfo | tr ' ' '\n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'
Containers use the CPU of the host (with Docker Desktop, that of the VM), so the host is what gets checked. In virtual machines, what counts is what the hypervisor passes through to the guest.
Which open-source applications are affected
The tables separate three levels: applications for which the discoverers report code execution, applications that merely run on Java, and projects that reacted as a precaution. "Affected" in the sense of "attackable" means for every application: unpatched JDK, a matching CPU and a path to the decoder. The vendor status applies as of 18 November 2026, 2 pm CET; "no statement found" means that no statement on Java4Shell could be found when checking project pages, advisories and repositories.
Reported as exploitable by Calif
Calif reports code execution in the default configuration. The chains are not published. According to The Hacker News, other labs have reproduced crashes and memory leaks, but not code execution.
| Application | Use | Path according to Calif | Origin of the JDK | Vendor status |
|---|---|---|---|---|
| Keycloak | Identity and access management | Before authentication | Container image: OpenJDK 21 from Red Hat UBI packages, images 26.8.2-1 and 26.8.2-2 with OpenJDK 21.0.13; operator image 26.8.2 with OpenJDK 25; ZIP and TAR.GZ: JDK of the host | Statement of 18 Nov.: may be affected, workaround through JAVA_OPTS_APPEND, no fixed image recommended yet |
| Jenkins | CI/CD | Before authentication | Docker image built from Temurin, in the repository on 18 Nov. Temurin 21.0.13+9; WAR: JDK of the host | No statement found; latest advisory of 28 Oct. |
| Elasticsearch | Search and logs | After authentication | Bundled JDK | No security notice; the bundled JDK was raised to 27.0.1.1 in the source code on 18 Nov. (pull request, backports to several branches planned), a release is pending |
| Apache Kafka (broker) | Messaging | After authentication | Docker image on Temurin 25 (JRE); tarball: JDK of the host | No statement found |
| Apache Solr | Search | After authentication | Docker image on Temurin 25; tarball: JDK of the host | No statement found |
| Gerrit | Code review | After authentication | External JDK (WAR) | No statement found |
| Spinnaker | Continuous delivery | After authentication | Not determined | No statement found |
In addition, Calif names the proprietary products Jira, Confluence, Artifactory and PingFederate (before authentication) as well as example applications based on Apache Struts and Spring Boot. According to Calif the list is not exhaustive.
Java-based, no statement on exposure
These applications run on the JVM. Whether they are attackable depends on the JDK, the CPU and the input paths. There is no statement from the discoverers or the projects.
| Application | Use | Origin of the JDK | Vendor status |
|---|---|---|---|
| OpenSearch | Search and analytics | Bundled, in the repository on 18 Nov. Temurin 25.0.5+8 | No statement found |
| Wazuh Indexer | Security analytics | Bundled (based on OpenSearch), JDK not determined | No statement found |
| Logstash | Log pipeline | Bundled | No statement found |
| Metabase | Business intelligence | Docker image on Temurin 25 (JRE); JAR: JDK of the host | No statement found; a pull request from a contributor is open |
| Apache Tomcat and applications on it | Servlet container | Docker image on Temurin 21 | No statement found; a researcher at Immersive names Tomcat Basic Auth as a path before authentication, according to Tenable a hint, not a tested result |
| XWiki | Wiki | Not determined | No statement found |
| ThingsBoard | IoT platform | Not determined | No reply to the open issue 16779 |
| Jitsi (Jicofo, JVB) | Video conferencing | Docker stack: Debian packages openjdk-21 | No statement found |
| Zabbix Java Gateway | Monitoring | Docker image: openjdk21-jre-headless (Alpine, Ubuntu) | No statement found |
| SonarQube, Sonatype Nexus Repository | Code analysis, artifacts | Not determined | No statement found |
Projects with a precautionary measure
Several projects adopted the flag into their delivery on 18 November. This is precaution and not a finding about the application:
- Dependency-Track sets
-XX:-UseBASE64Intrinsicsin the product (pull request 8037). - Camunda 8 protects the distributions because Base64 is decoded from untrusted input (pull request 69744).
- openHAB sets the flag by default in the distribution (pull request 2004).
- Graylog adopted the flag into the package build (pull request 155).
- Apache Roller sets
CATALINA_OPTSin the Docker image (pull request 311).
Where your application's JDK comes from
Who has to patch depends on where the JDK comes from. The same application can appear in different rows depending on the type of installation.
| Origin | Examples | Who delivers the fix | What to do |
|---|---|---|---|
| JDK package of the operating system | Jitsi Docker stack (Debian), Zabbix Java Gateway, ZIP and WAR installations | Distribution | Apply the package update as soon as it is available; use the workaround until then |
| Base image with Temurin | Jenkins, Kafka, Solr, Metabase, Tomcat image | Image vendor after the Temurin release | Pull the new image or rebuild it yourself, recreate containers |
| Vendor image with its own JDK selection | Keycloak (Red Hat UBI packages) | Vendor | Workaround until a confirmed fixed image is released |
| Bundled in the product | Elasticsearch, OpenSearch, Logstash | Vendor with a release | Workaround, wait for the release; replace the JDK only as specified by the vendor |
| Appliance or third-party product | Software with a built-in JVM | Vendor | Apply the vendor update, clarify interim measures with the vendor |
The version number of an application or an image tag does not by itself show which JDK it contains. What gets checked is the running JVM, not the Java in the shell path (Datadog).
Patch status of the JDK vendors
Oracle names 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1 as fixed versions, and 8u513-perf for the Java 8 Performance Pack; other vendors assign their own build numbers. The vendors' status as of 18 November 2026, 2 pm CET:
| Vendor | Status | Source |
|---|---|---|
| Oracle Java SE | Patched since 17 Nov. | Oracle |
| Amazon Corretto | Patched: 17.0.21.1.2, 21.0.13.1.2, 25.0.5.1.2, 27.0.1.1.4; the changelog names the CVE | Corretto 21 changelog |
| Azul Zulu | Patched with the November CSPU, Prime 26.08.110.0 and 26.02.510.0 | Azul |
| Eclipse Temurin | Fixed versions 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1 named; binaries and Docker images follow after testing | Adoptium |
| Red Hat OpenJDK and RHEL | Affected, fixes in preparation, no errata yet | RHSB-2026-018 |
| Debian | Only sid fixed (openjdk-17, 21, 25, 27); bookworm, trixie and forky listed as vulnerable | Security Tracker |
| Ubuntu | openjdk packages of the listed releases at "Needs evaluation", no fix | Ubuntu Security |
The reports partly contradict each other: Calif and several trade outlets initially reported no Temurin fix, whereas the Adoptium post of 18 November names fixed versions. The version numbers are therefore confirmed; the availability of the binaries and Docker images depends on the ongoing release. The assessments of the severity also differ: 8.1 at Oracle and in the OpenJDK advisory, 9.2 at Calif, provisionally 9.8 at Red Hat.
Workaround: switching off the Base64 intrinsic
Until a patched JDK is running, the option -XX:-UseBASE64Intrinsics switches off the vulnerable path. Red Hat, Keycloak and Adoptium name it as the preferred interim solution: it disables only the accelerated Base64 routine. Other AVX-512 optimizations stay active. The alternative -XX:UseAVX=2 switches off AVX-512 completely and may only be set on x86-64, because JVMs on ARM64 do not start with it.
How the flag gets into the JVM depends on the application:
| Application or environment | Method | Source |
|---|---|---|
| Keycloak (bare metal, Docker, Compose) | JAVA_OPTS_APPEND; with systemd in the service environment, not in the interactive shell |
Keycloak |
| Keycloak Operator | In the Keycloak resource spec.env or spec.unsupported.podTemplate; for the operator itself JDK_JAVA_OPTIONS |
Keycloak |
| Jenkins (Docker image, controller) | JENKINS_JAVA_OPTS, the variable for JVM options documented in the Docker README |
Jenkins Docker README; the flag recommendation comes from Red Hat, Jenkins has not commented |
| Tomcat | CATALINA_OPTS; in the Apache Roller image, a value passed through Docker replaces the default |
Apache Roller |
| Apache Flink | env.java.opts.all in the Flink configuration |
Draft of the Flink blog, not yet published (pull request 896) |
| Any JVM | JAVA_TOOL_OPTIONS is evaluated by HotSpot JVMs at startup; Start9 uses it for Jicofo and JVB |
Start9 package for Jitsi |
| Elasticsearch, OpenSearch | The usual way to pass additional JVM options is ES_JAVA_OPTS or OPENSEARCH_JAVA_OPTS; there is no vendor recommendation on Java4Shell |
Vendor documentation on JVM options |
Prove the effect. What counts is the running JVM, not the variable that was set (Keycloak):
jcmd <pid> VM.flags -all | grep UseBASE64Intrinsics
# expected: UseBASE64Intrinsics = false {command line}
docker exec <container> jcmd 1 VM.flags -all | grep UseBASE64Intrinsics
If an image does not contain jcmd, the remaining option is to check the process command line with ps or docker inspect; this shows what was started but does not reliably prove the effective value. After every restart and rollout the check has to be repeated for every process, including failover nodes and nodes in a rolling update.
What to do now
- Take inventory. Which applications run on Java, on which hosts, in which containers and appliances? The running processes count, not just the installed packages.
- Clarify the JDK origin. Record for each application whether the JDK comes from the operating system, the base image, the vendor or the product. This determines who delivers the fix.
- Check the CPU of the hosts. Hosts without AVX-512 VBMI and ARM64 hosts need no measure. For all others it depends on whether the JDK is unpatched and whether a path to the decoder exists.
- Prioritize. First applications reachable from the internet, then central authentication services and business-critical systems (BACS, Red Hat). Keycloak and Jenkins belong at the top.
- Set the workaround, restart, verify. Set the flag per application, restart the service and prove the effect with
jcmd. - Update the JDK. Apply packages, pull base images again and rebuild application images, recreate containers. Then check the JDK version that is running.
- Follow vendor releases. For Keycloak, Elasticsearch, OpenSearch and other applications with their own JDK, watch the vendor's notice and apply the update as soon as it is confirmed.
- Monitor. Watch for unusual JVM crashes, unexpected processes and suspicious activity. An update does not remove a compromise that has already happened (BACS).
Is the flaw being exploited?
As of 18 November 2026 no exploitation in the wild is confirmed, and the flaw is not in the CISA Known Exploited Vulnerabilities Catalog. A reproducer from Calif is public, which triggers the memory overwrite and checks whether the Java version and CPU are affected. Calif is withholding the chains for individual products until operators have had time to update and intends to publish them later. A researcher at Immersive reports on X a working application-specific exploit that is not published (Tenable). The cause and the trigger are public. Anyone operating authentication services and build systems should therefore not wait for the chains to be published.
Timeline
| Date | Event |
|---|---|
| 17 November 2026 | Oracle publishes the November update, the OpenJDK project its advisory with ten fixed flaws, including CVE-2026-114938; Calif publishes details that, according to Tenable, had leaked beforehand |
| 17 November 2026 | Amazon Corretto and Azul publish patched builds; Red Hat publishes RHSB-2026-018 without a fix |
| 18 November 2026 | Keycloak publishes notes and workaround; Adoptium names fixed Temurin versions |
| 18 November 2026 | Dependency-Track, Camunda, openHAB and other projects adopt the flag; Debian sid receives patched JDK versions, and Elasticsearch raises the bundled JDK in its source code |
Common misconceptions
- "My server runs on AMD, so I am safe." No. AVX-512 VBMI is offered by AMD EPYC and Ryzen from Zen 4 onwards. What counts is the CPU feature, not the manufacturer.
- "My application is up to date, so I am safe." No. The JDK decides. Keycloak explicitly points out that the version of the application alone says nothing about the fix status.
- "The flag is set, so that's done." No. The service has to restart, the effect has to be checked on the running JVM, and Red Hat describes the workaround as an interim measure only.
- "I have to patch Java 8 and 11 as well." According to the OpenJDK advisory and Adoptium they are not affected by default. The exception is Oracle's variant 8u511b33-perf.
- "The flaw only affects Atlassian and other products." No. With Keycloak and Jenkins, Calif also names open-source applications, and the cause lies in the JDK that every Java application brings along or uses.
- "This is like Log4Shell." That cannot be demonstrated. There is no universal exploit, each application needs its own chain, and the preconditions are narrower. Given the reach into authentication services, prompt securing remains the right call.
Status and methodology
As of 18 November 2026, 2 pm CET, a few hours after publication. The technical information comes from the OpenJDK advisory, the Calif publication, the Red Hat bulletin, the Keycloak notice, the Adoptium post, the Azul post, the trackers of Debian and Ubuntu, the Corretto changelog and the reports by Datadog, Tenable, Qualys, The Hacker News, heise online and the BACS.
The information on the origin of the JDK comes from the projects' official repositories and Dockerfiles, as of 18 November. Where "not determined" appears, the JDK was not checked. "No statement found" describes the result of reviewing project pages, advisories and repositories; mailing lists and support portals requiring a login could not be searched. Statements on the exploitability of individual applications rest on Calif's information and are not independently confirmed.
Our approach at WZ-IT
- Inventory: We record where Java runs in your environment, where the JDK comes from and which hosts offer AVX-512 VBMI.
- Securing: We set the workaround per application, restart the services and prove the effect on the running JVM.
- Patching: We apply JDK updates, rebuild container images on a patched base and verify the running version.
- Reviewing: We examine logs and JVM crashes for anomalies and rotate credentials if a suspicion arises.
- Operations: In managed operations, updates, CVE monitoring and backups are included; the server audit reviews existing systems once.
Further guides
- Why companies need CVE monitoring
- PHP flaws CVE-2026-109618 and CVE-2026-109619: affected software and measures
- SSHole CVE-2026-123158: checking Alpine images with sshd
- Keycloak 26.8: upgrade guide
- OpenSearch or Elasticsearch: comparing vendor lock-in
- Metabase: SQL injection with CVSS 10 and emergency patch
- Keycloak operations by WZ-IT
- CVE monitoring
Check your Java estate without finding out by chance: We record where Java runs in your environment, set the workaround and apply the updates as soon as they are available. Book an initial consultation
Sources
- Calif: Java4Shell
- Oracle: Critical Security Patch Update November 2026
- OpenJDK Vulnerability Advisory 2026/11/17
- Red Hat: RHSB-2026-018 Remote Code Execution via memory corruption on x86-64 systems
- Keycloak: Java4Shell - What Keycloak admins need to do
- Adoptium: Temurin November 2026 CSPU and Java4Shell
- Azul: Java4Shell - what it is, who is affected and what to do now
- Debian Security Tracker: CVE-2026-114938
- Ubuntu Security: CVE-2026-114938
- Amazon Corretto 21: Changelog 21.0.13.1.2
- Datadog Security Labs: Java4Shell
- Tenable: Java4Shell FAQ
- Qualys: How to Find, Mitigate, and Patch the Critical Java4Shell Flaw
- The Hacker News: Java4Shell Flaw Enables Pre-Auth RCE
- heise online: Java4Shell - Gefährlicher Bug in Java (in German)
- BACS: Kritische Schwachstelle in Java (in German)
- CISA: Known Exploited Vulnerabilities Catalog













