Java4Shell lets attackers into Jira, Jenkins and Keycloak before login. Find what runs Java and patch it tonight
A bug in Java's Base64 code, disclosed today, gives attackers a way into Jira, Jenkins, Keycloak and custom Spring apps before anyone logs in. Here's how to find the Java you didn't know you ran, and what to do with it tonight.

Java4Shell (CVE-2026-114938), disclosed November 17, 2026, is a memory bug in OpenJDK's fast Base64 decoder. It affects Java 17.0.6 and later, 21, 25 and 27, but only on x86-64 processors with AVX-512 VBMI; ARM is not affected. Install your vendor's fixed Java (Oracle: 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1). Until it ships, restart with -XX:-UseBASE64Intrinsics, then confirm the running process uses it.
What happened this afternoon
A security firm called Calif published details today, November 17, of a bug it calls Java4Shell (CVE-2026-114938). It sits in a small, very fast piece of Java's own code that decodes Base64, the format software uses to squeeze login tokens, file uploads and network messages into plain text. Send that code a slightly broken string, and on the right kind of processor it writes past the end of its memory. Calif's trigger is 64 letter A's followed by an equals sign. Calif's write-up shows it turning that into a full takeover of the server.
Calif says it got remote code execution, before anyone logs in, on default installs of Jira, Confluence, Jenkins, Keycloak, JFrog Artifactory and PingFederate. With a valid login it got the same result on Kafka, Solr, Elasticsearch, Gerrit and Spinnaker. It also broke sample apps built on Spring Boot and Apache Struts through features as ordinary as sign-in and file upload. A separate exploit takes over a Java program that merely connects to a malicious server over TLS. Calif is holding back the product-by-product exploits for now. The bug itself is public, and Calif says it had already leaked before the patch.
Oracle shipped fixed Java builds at 1:00 PM Pacific today. As of this afternoon nobody had reported attacks in the wild. Calif says finding the bug took one of its researchers an hour with AI help, and building the first working exploit took two more. If it took them three hours, assume other people can do it in a few days. That's why this is a tonight job, not a next-sprint job.
Do you run Java without knowing it?
Ask most owners and they'll say no. Then you list the tools, and Java shows up in half of them. Java is rarely the product you bought. It's the engine underneath. Go down this list and mark each one as "we host it", "a vendor hosts it" or "don't know":
- Atlassian you host yourself. Jira or Confluence Data Center on your own servers or cloud account. (Atlassian's own cloud is Atlassian's job.)
- Jenkins, or anything else that builds and deploys your code: the box under someone's desk counts.
- Your sign-in. Keycloak, PingFederate, or a "log in with the company account" page your developers set up.
- Search and data plumbing. Elasticsearch, Solr, Kafka, often added years ago for site search or event feeds.
- Your own backend. If your app was built in Spring Boot, Kotlin or Scala, it runs on Java. Ask whoever built it.
- Appliances and installed software. Many vendor products ship with their own copy of Java inside, and the Java on the server doesn't tell you what's in the product folder.
- Managed services. Hosted Elasticsearch, hosted Kafka, a SaaS sign-in provider. Your vendor patches these, but you still want it in writing.
The "don't know" rows are the work. The riskiest machine is usually the one nobody has opened in two years. We wrote about the server you stopped using for exactly this reason.
Java is rarely the product you bought. It's the engine underneath.
Two checks tell you how worried to be
Java4Shell needs two things at once, which makes it easier to triage than the headlines suggest.
The Java version. The affected code is in OpenJDK 17.0.6 and later, and in Java 21, 25 and 27. The OpenJDK advisory marks Java 8 and 11 as unaffected by this CVE. That covers every build made from OpenJDK, which is nearly all of them: Temurin, Amazon Corretto, Azul Zulu, the Microsoft and Red Hat builds, Liberica, SapMachine, Oracle's own Java SE and GraalVM. Run java -version inside the container or on the server that actually runs the app, not on your laptop.
The processor. The buggy code only runs on x86-64 chips with a feature called AVX-512 VBMI. Calif names Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids and Granite Rapids, and AMD Zen 4 or newer. Those are the chips behind a lot of current cloud machines; Upwind lists AWS c6i, m6i, c7i, c7a and m7a. ARM servers, like AWS Graviton, are not affected. On Linux, Calif's check is:
grep -m1 '^flags' /proc/cpuinfo | tr ' ' '\n' | grep -E '^(avx512f|avx512bw|avx512vbmi|bmi2)$'
If all four come back, the machine can run the vulnerable path. One warning: cloud fleets mix chip generations, and a server can move to newer hardware on its next restart. A "no" today is not a reason to skip the patch.
How the work moves
Is this machine exposed? A five-minute triage
- Does it run Java?
Check the app, the container image and any Java bundled with the product.
- Which version?
17.0.6 or later, 21, 25 or 27 is in scope. Java 8 and 11 are not, for this CVE.
- Which processor?
x86-64 with avx512vbmi in /proc/cpuinfo is exposed. ARM is not.
- Who can reach it?
Public sign-in pages and anything taking uploads go first.
- Patch or flag, then restart
Fixed vendor build, or -XX:-UseBASE64Intrinsics until it ships.
What to do tonight, by where it runs
Servers and VMs you manage. Install the fixed Java from whoever supplies yours. Oracle's builds are 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1. As of this afternoon, Temurin, Corretto and Zulu fixes were still pending, and Linux distributions hadn't shipped packages yet. If your build isn't out, don't wait for it. Add -XX:-UseBASE64Intrinsics to the Java startup options and restart. That switches off the fast decoder, and Upwind found it stopped the overflow on every affected version it tested. The cost is slower Base64 decoding, a fair trade for one night.
The other flag, carefully. -XX:UseAVX=2 also works, but only on x86 machines. Calif warns that Java on ARM doesn't recognize it and won't start. If one startup script is shared by mixed hardware, use the Base64 flag. Also skip the -XX:DisableIntrinsic=... line going around: on its own it stops the JVM from starting.
Containers. The Java version is baked into the image. Rebuild on a patched base image, or add the flag through your deployment settings, then redeploy. Restarting the old image changes nothing.
Atlassian, Jenkins, Keycloak and other packaged products. Many of these bundle their own Java. Patch the bundled one or set the flag where the product reads its startup options, and check the vendor's advisory page for its own fixed release. Put internet-facing sign-in pages and anything with a public login first.
Managed and SaaS. You can't patch these. Send the questions below.
Every option here means a restart. Do it like any risky update: know how you'll roll back and who checks the app afterwards. Same as our WooCommerce 11.3 note, just with a shorter clock.
Decision guide
Tonight's move by where Java runs
| Consideration | Do tonight | Then |
|---|---|---|
| Your own servers | Install the fixed build if your vendor has one; otherwise add -XX:-UseBASE64Intrinsics and restart. | Swap the flag for the fixed build when it ships. |
| Containers | Rebuild or redeploy with the flag; restarting the old image does nothing. | Move to a patched base image. |
| Packaged products | Set the flag where the product reads startup options; public logins first. | Install the vendor's fixed release. |
| Managed and SaaS | Send the three vendor questions. | Keep the written answer with your list. |
Three questions for every vendor that runs Java for you
Send these tonight to whoever hosts your sign-in, search, build server or any product on your "vendor hosts it" list. A good vendor answers within a day.
- Are you exposed to CVE-2026-114938 (Java4Shell), and on which systems? You want a yes or no per service, not a statement about how seriously they take security.
- What have you done, and when? Patched Java to which version, or set
-XX:-UseBASE64Intrinsics, with the date and time. "We're monitoring" is not an answer to this one. - Did anything reach it first? Ask whether they've checked logs for malformed Base64 errors and unexpected JVM crashes since the details went public. That's the honest version of "were we hit?"
If a vendor can't answer the first question by tomorrow, treat the service as unpatched and limit who can reach it until they do.
How to know you're actually done
Patch notes are easy to send and easy to misread, so check the running thing.
- Every row on your list has an owner and a status: patched (with version), flagged (with restart time), not affected (with the reason: ARM, Java 11, no AVX-512 VBMI) or vendor-confirmed in writing.
- The running process shows it.
java -versioninside each container or host shows the fixed build, or the process's startup command includes the flag. A config file nobody restarted doesn't count. - Somebody looked at the logs. In Upwind's tests, every attempt left
IllegalArgumentException: Input byte array has wrong 4-byte ending unitin the application logs. Unexplained JVM crash files (hs_err_pid*.log) or bursts of restarts since this afternoon are worth a closer look. - Flags have an end date. The flag is a stopgap. Put a reminder on the calendar to swap it for your vendor's fixed build when it ships, then take the flag out.
If your list has more "don't know" than you'd like, or the app that matters most was built by a team that has since moved on, that's work we do. TWIMCO builds and looks after web applications, including the Java and Spring backends and sign-in systems this bug reaches. Tell us what you're running and we'll help you sort tonight's list into patched, flagged and confirmed.
Related work: PatientIQ.
We build alongside the people doing the work.
Bring us the problem you’re thinking about.