Java4Shell (CVE-2026-114938): Heartbleed All Over Again, This Time in Java

On November 17, 2026, the researchers at Calif disclosed Java4Shell (CVE-2026-114938): a bug in the Base64 decoder that ships inside modern Java itself. In their tests it gave an attacker with no account remote code execution (RCE) on Keycloak, Jira, Confluence, Jenkins, Artifactory and PingFederate.
Those products are only where it was demonstrated. The flaw is in the Java runtime, not in any one product.
The same bug also lets an attacker read the memory of the server process. Anything held there can leak out: private keys, passwords, session tokens and other secrets.
This article covers what we know, how to mitigate it and how to validate that mitigation with 10 new attack scenarios in Cymulate Exposure Validation that test and validate your WAF and IPS to block or detect attacks that target the Java4Shell vulnerability.
As VP of research at Cymulate, I’m incredibly proud of how the Cymulate Research Labs team rapidly developed and released these 10 new production-safe attack simulations within 12 hours of Java4Shell’s public disclosure.
What is Java4Shell (CVE-2026-114938)?
Java4Shell is a heap overflow in the fast Base64 decoder that OpenJDK uses on x86-64 CPUs with AVX-512 VBMI. A short, badly padded Base64 string makes the decoder write about 62 bytes past its buffer. Java rejects the input as invalid, but only after the write has already happened. The trigger is tiny. Calif’s example is 64 A characters followed by a single =, 65 bytes in total.
| Condition | Affected | Not affected |
| Java version | OpenJDK 17.0.6 and later, including 21, 25 and 27, and builds based on it (Temurin, Corretto, Zulu, Red Hat, Microsoft and others) | Standard Java 8 and Java 11 (Oracle Java SE 8 Enterprise Performance Pack is affected) |
| CPU | x86-64 with AVX-512 VBMI: Intel Ice Lake to Granite Rapids, AMD Zen 4 and newer | ARM64, and x86-64 without AVX-512 VBMI |
| Fixed in | 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1 |
Oracle scores it 8.1 under CVSS 3.1. Calif scores it 9.2 under CVSS 4.0.
Why is Java4Shell so dangerous?
The bug sits in Java itself, so any Java service that decodes Base64 from a request can be exposed. Servers decode Base64 all the time, and often before they check who you are.
- No login needed. Basic auth headers, tokens, SAML messages and file uploads are all decoded before authentication. Calif reports pre-auth RCE on Keycloak, Jira, Confluence, Jenkins, Artifactory and PingFederate, and post-auth RCE on Kafka, Solr, Elasticsearch, Gerrit and Spinnaker.
- The worst targets come first. Identity providers, build servers and artifact stores hold the keys to everything else. One compromised Keycloak or Jenkins is a path into the whole environment.
- It also leaks memory. The same corruption gives a Heartbleed-style read of the heap, which Calif calls JavaBleed. That memory can hold private keys, session tokens and passwords.
- Clients are exposed too. Calif says a Java client that connects to a malicious TLS server can be compromised.
- The long tail is long. Many products bundle their own JDK inside the app or the container image. Patching the operating system does not fix those copies, so vulnerable Java will run quietly for years.
How to mitigate Java4Shell: patches and temporary workaround
- Patch the JDK, not just the OS. Move to 17.0.21.1, 21.0.13.1, 25.0.5.1 or 27.0.1.1, or your vendor’s build that names CVE-2026-114938. Check every app and container image that bundles its own Java.
- Temporary workaround: turn off the vulnerable decoder. Where you can’t patch yet, start the Java Virtual Machine (JVM) with
-XX:-UseBASE64Intrinsics. The minus sign switches the option off, so Java stops using the hand-written AVX-512 Base64 code where the bug lives and falls back to its normal decoder.- It only takes effect after the JVM restarts with the flag.
- Check the running process, not just the config:
jcmd <pid> VM.flags -all | grep UseBASE64Intrinsicsshould showfalse. - Expect some slowdown in apps that decode a lot of Base64. Test before you roll it out widely.
-XX:UseAVX=2also closes the hole on x86 by turning off AVX-512 entirely, but it costs more speed and Java will not start with it on ARM.- Either flag is a stopgap. Remove it once the host runs a fixed build.
- Patch in order of blast radius. Internet-facing identity providers first, then Jira, Confluence, Jenkins and Artifactory, then everything else that handles logins, tokens or uploads.
- Shrink the exposure. Put consoles that don’t need the internet behind a VPN until they are patched.
- Know your detection gaps. Talos has Snort rules for subscribers, community Snort and Suricata rules exist, and WAF vendors ship interim rules. Generic rule sets will not catch a short Base64 string, and nothing on the network sees inside HTTPS unless you decrypt it.
- Watch the hosts. Unexpected JVM crashes (
hs_err_pidfiles), Base64 decode errors in logs, and Java processes spawning shells or making new outbound connections are your best late signals.
A personal note: Black October was only the beginning
October felt like a storm. Critical bugs came one after another, each with its own “2Shell” name, and attacks followed fast. Many of us called it Black October and hoped it was the peak.
I don’t think it was. I think it was the opening act, and the real aftermath is still ahead of us. Java4Shell, arriving weeks later in one of the most widely used enterprise runtimes, is a reminder of that.
What worries me most are the bugs nobody publishes. Calif disclosed this one responsibly, waited for a patch and held back its exploits. Not everyone will. Some will keep a bug like this, sell it, or use it unnoticed for months.
And then there’s AI. Finding a bug like this was never the slow part. Turning it into a working exploit for each product was. That took skilled people and weeks of trial and error. Now a few good people with the right tools can get through that work much faster, and from the outside you can’t tell who is doing it.
There’s also a lot more noise than before: fake PoCs, copied advisories, alerts that lead nowhere. Defenders have to dig through all of it. My guess is that the time between disclosure and exploitation will keep getting shorter, and that some of these bugs are already in use before anyone writes about them.
We don’t know what we don’t know. Plan as if an attack can come at any time, from anywhere, through something you haven’t seen yet.
How do you test your defenses against bugs like Java4Shell?
You can’t patch a bug nobody has disclosed. But you can know, with proof, how your defenses react when something new hits them. That is what exposure validation is for, and it only works if you run it often, not once a year.
For a bug like Java4Shell, a safe simulation sends the trigger between test machines and asks a few simple questions:
- WAF and IPS/IDS: Did they see the request and block it?
- SIEM: Did the event arrive, and did it become an alert?
- EDR: If a Java process crashes or starts a shell, does anyone find out?
- SOC: Does the team know what the alert means and what to do in the first hour?
- Compensating controls: If you can’t patch yet, does your workaround really hold?
Every gap you find is a gift. It shows you where a control stops working, so you can fix it before an attacker finds it. That is how you close the risk-to-fix gap: prove which controls miss the attack, prioritize the gaps that matter and adapt your defenses with updated detection rules, IoCs and control settings. Then test again after every change, because what worked last month can break this month.
Java4Shell will be patched, slowly, like every big bug. The next one will not wait. Patch what you can this week. Then go and prove that your defenses work, and update them where they don’t.
New Cymulate attack scenarios for WAF and IPS
To answer the first of those questions, Cymulate has released 10 Java4Shell scenarios. Each one sends a broken Base64 value that triggers the decoder bug:
- 5 WAF scenarios: send the request to a web app to test whether your WAF blocks it.
- 5 Traffic simulation scenarios: replay the same five requests as network traffic on port 80 between two Cymulate agents to test whether your IPS/IDS detects them.
If a control misses the attack, you know exactly which rules to fix.
Already a Cymulate customer? Run the Java4Shell scenarios now.
Not using Cymulate yet? Reach out and we’ll show you how to run the same test in your environment. See it in action.
Java4Shell FAQ
As of November 17, 2026, Qualys and Upwind reported no confirmed exploitation in the wild, and the CVE was not in CISA’s KEV catalog.
OpenJDK 17.0.6 and later, including 21, 25 and 27, and builds based on it, but only on x86-64 CPUs with AVX-512 VBMI. Fixed builds are 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1.
Yes. Restart the JVM with –XX:-UseBASE64Intrinsics to turn off the vulnerable decoder. It is a stopgap until you patch, and it can slow down apps that decode a lot of Base64.
Log4Shell was a bug in a logging library. Java4Shell is in the Java runtime itself, so a service can be exposed even if none of its libraries are at fault.